Physics for health

For protons and 3He/4He similar radio-biological properties have been determined, but the lateral scattering is reduced by nearly 50% in the case of helium ions versus protons (figure 5.7). In recent years, helium ions again became of interest for clinical cases where neither protons nor carbon ions are ideally suited, especially for treating paediatric tumours. Currently, patient irradiations with scanned ⁴He ions at the Heidelberg Ion beam Therapy Center (HIT) in Germany are used only in 'treatment attempts' ('individuelle Heilversuche') and will go into regular operation by mid 2024. Other ion therapy facilities in Europe (e.g., CNAO in Italy and MedAustron in Austria) have also started technical upgrades to produce helium ion beams in the near future.

Zoom In Zoom Out Reset image size

Recent studies have shown that 3He ions can be a viable alternative to ⁴He, as they can produce comparable dose profiles, demanding slightly higher kinetic energy per nucleon, but less total kinetic energy. This results in 20% less magnetic rigidity needed for the same penetration depth which may be of importance for the design of future compact therapy accelerators like superconducting synchrotrons or energy-variable cyclotrons.

The observation of the patient's position during treatment has become more and more a standard procedure. 3D camera systems make it possible to control the location of the patient within tenths of millimetres observing the exterior of the body. But organs struck by tumours can move dramatically within the abdomen, which is not visible from outside. To observe the soft tissues an MRI scanner is the preferred diagnostic tool. But due to the moderate to high fields used the charged particle beam is affected. In addition, conventional MRI scanners are not constructed to have an inlet for radiation to be applied in parallel to the diagnostic procedure. Nevertheless, studies like ARTEMIS at HIT in Heidelberg have been started to investigate possible arrangements of open, low-field MRIs (figure 5.8). Still the deflection and/or distortion of the beam can be observed, but algorithms will be developed to cancel out this influence of the MRI's magnetic field. A second goal of such studies is to optimize the design of the measurement coils to spare out space for the beam entrance fields. A further aspect should not be underestimated—the impact of the magnetic (stray) fields on the QA devices and the online monitoring detectors used during treatment. All these have to be examined and possibly adapted to be magnetic field compatible; at worst new or alternative detector principles have to be applied. A lot of technical solutions have to be found in the next years before MRI scanners can be introduced in regular particle beam therapy.

Zoom In Zoom Out Reset image size

To enhance the coverage of particle therapy in Europe and worldwide and to enlarge the number of patients that can profit from this special treatment, the investment costs of such facilities should be reduced as much as possible, which requires smaller and simpler machines to reduce manufacturing and operating effort. Especially the size of the accelerator has an important influence on the building costs. And the amount of beam losses demands more or less concrete for radiation shielding, and should be minimized by design. In addition, the operating and maintenance team needed should be small, but adequate and well-trained, and sustained by a modern control system, which predicts pre-emptive maintenance measures through AI algorithms and thus guarantees highest availability.

Energy-variable cyclotrons

Most proton therapy facilities use cyclotrons to produce the beam, as these accelerators are compact, have only a few tuning parameters, and are thus simple to operate. Over the last three decades the size and weight have dramatically shrunk by factors 3–10 while the magnetic fields using superconductivity were increased up to a factor 4 (figure 5.9). These improvements still show the potential of cyclotrons while parallel developments like proton linacs are still larger and have a much more complex technique. But the actual cyclotron generation consists still of fixed energy machines, which demands a degrader for energy variation causing two main drawbacks: (a) high local beam losses and (b) relatively low currents for mid-depth and skin-deep tumours, which may be a problem to use the FLASH-based treatment procedure. To overcome these disadvantages first studies of energy-variable cyclotrons were undertaken and published. These show the possibilities to consequently advance the cyclotron systems to lighter, high-field superconducting arrangements, but now iron-free and thus capable to vary the magnetic field and energy in reasonable times. In addition, the produced currents of such a setting do not depend on the energy because no loss mechanism is involved anymore. The resulting machines would be 'FLASH'-ready and have the additional possibility to be enhanced to ³He/⁴He beam combined with proton therapy in one cyclotron setup. More simulation studies and prototype constructions are needed to reach these goals in the next 5–10 years.

Zoom In Zoom Out Reset image size

Superconducting fast-ramped synchrotrons

The first proton and ion linac-based facilities

Towards a VHEE RT facility

NC RF linac is the technology being used for most of the VHEE research. The main advantages of the linacs are the flexibility and the compactness. Regarding the linac design in the energy range of interest for VHEE applications there are different possibilities offering the desired performances and compactness with different degrees of technology maturity. The S-band technology is the most mature one; HG compact linacs of this type are already available from various industrial partners. The C-band and X-band RF linacs are still less mature and are mainly constructed in labs with the help of industries for machining. Recently a considerable effort is being made from the industrialization point of view. The current and next future available machines for VHEE are the eRT6-Oriatron at Centre Hospitalier Universitaire Vaudois in Laussane; ElectronFlash at IC in Orsay; CLARA at Daresbury; AWA at ANL; and CLEAR at CERN (all based on NC RF linacs). A VHEE-FLASH facility based on a CLIC X-band 100 MeV linac is being designed in collaboration with CHUV to treat large, deep-seated tumours in FLASH conditions. The facility is compact enough to fit on a typical hospital campus (figure 5.5). Another proposal in this sense is the upgraded PHASER proposal at SLAC. Finally, ELBE at HZDR and the next future PITZ at DESY are based on SCRF linac technology.

Zoom In Zoom Out Reset image size

Recent advances in the high-gradient RF structures where more than 100 MeV m⁻¹ are now achievable in the lab environment are transforming the landscape for VHEE RT. VHEE RT requires beam energies between 50 and 200 MeV, an improved dose conformity and scale to higher doses rates, in the case of the FLASH-RT until 50 Gy s⁻¹ are needed. Novel high-gradient technologies could enable ultra-compact structures, with higher repetition rates and higher currents. An international R&D global effort is being made by major accelerator laboratories and industry partners and is focused on two aspects: material origin and purity, surface treatments, and manufacturing technology on the one hand and the consistency and reproducibility of the test results on the other. Some promising R&D in the next decade are the distributed coupling accelerator developed at SLAC and the use of cryogenic copper that is transforming the linac design offering a new frontier from beam brightness, efficiency, and cost capability. Another approach for the next generation of compact, efficient, and high-performance VHEE accelerator is the use of higher-frequency millimetric waves (∼100 GHz) and higher repetition rates using THz sources.

An important R&D effort to apply these technologies in the medical industry has to be made in the next decade, if successful, this could be a step further in the quest for compact and efficient VHEE RT in the range of hundreds of MeV. For achieving these aims, a synergistic and multidisciplinary research effort based on accelerator technology as well as physical and radiobiological comparisons to see how well VHEE can meet the current assumptions and become a clinical reality is needed (Very High Energy 2020).

Therapy facilities based on laser plasma acceleration

Robotic systems transform the way we work and live, and will continue to do so in the future. At the macroscopic scale, robots greatly speed up and enhance manufacturing processes, provide assistance in diverse areas such as healthcare or environmental remediation, and perform robustly in environments that are inaccessible, too harsh, or too dangerous for humans. In the lab they can, in principle, perform large numbers of experiments in parallel, reproducibly and without getting tired. This supports, for example, the search for new chemical compounds in combinatorial approaches, and can also generate the large datasets required by data-hungry machine learning techniques.

Robotic systems are 'reprogrammable multifunctional manipulators' and typically comprise sensors and actuators connected to and coordinated by an information-processing unit. Sensors provide information about the environment, which is evaluated by a computer and then used to decide on the necessary actions—which often means mechanical motion of some sort. At a macroscopic scale, a wide range of sensors, electromechanical components, and powerful—potentially networked—electronic computers are available to realize robotic systems, which are essentially all powered by electricity. Is it possible to realize robotic functions also at the micro- or even nanoscale, where we have to work with molecular components, and on-board electronics is not available?

In fact, over the past years researchers have begun to work on the development of 'molecular robotic systems', in which sensors, computers, and actuators are integrated within molecular-scale systems. Among the many possible applications for molecular-scale machinery and robots are, most prominently, the generation of nanomedical robots that autonomously detect and cure diseases at the earliest stages, and the generation of molecular assembly lines that will enable the programmable synthesis of chemical compounds.

Biology as a guide for molecular and cell-scale robotics

From molecular machines to robots

DNA-based robots

Active systems

Collective dynamics and swarms

Cells as robots

Power supply

Hybrid systems

Applications envisioned for micro- and nanorobotics

In the sections above, a wide range of challenges and opportunities associated with the development of future bioinspired micro- and nanorobots were mentioned, which are summarized more concisely here. A variety of challenges relate to the way robotic functions will be implemented in the first place:

• Some robotic functions will be realizable with nanoscale systems (composed of supramolecular assemblies, DNA structures, biological motors); others will require larger cell-like systems into which multiple functions are integrated. It will be important to understand the size dependence of these functions and clarify what can be achieved at which length scale, and with which components. Systems integration—combining the components into consistent and functioning systems—will be the major challenge for bioinspired small-scale robotics.
• Autonomous behavior versus external control. How complex do robotic systems need to be to act autonomously—and is autonomy required? For many applications, external control will be a more feasible and powerful approach, which also benefits from established macroscale technologies.
• Hybrid robots that combine the advantages of biological systems and traditional robotics seem promising—a major challenge is the realization of efficient interfaces for signal transduction and actuation between biological and conventional robotics.
• Robotic systems should perform useful tasks, which means they need to be operated under realistic conditions. Depending on the field of application, various practical issues will have to be tackled—operation inside a living organism or on the surface of a sensor chip come with very different requirements.

Many physical challenges arise from the question of whether we can create something like a robot at such small scales at all:

• Can small robots sense and respond to small numbers of molecules or photons? There are biological examples of extremely sensitive sensors that require special molecular architectures and amplification processes, which may guide the design of microrobotic sensors. A related question that has been studied extensively in biology is chemotaxis, where cells sense the presence of a concentration gradient.
• What forces can be usefully applied by small robots? Biological molecular motors are known to generate piconewton forces, and the same magnitude is also expected from artificial motors. However, synthetic molecular motors have not yet been used to perform any useful task.
• How fast can a nano- or microrobot move or act? Do we need to achieve directional movement or will diffusion be sufficient? Of course, this again will depend on the application. Interestingly, in biology we do not find molecular motors for transport of molecules in bacteria (diffusion is sufficient), but in the much larger eukaryotic cells. Further, the smallest bacteria tend to be non-motile—apparently motility only pays off above a certain size. There are various further issues such as the dominance of Brownian effects at the nanoscale.
• How much energy is required to perform robotic functions—and how will it be supplied? Should nano- or microrobots be autonomously driven by chemical reactions, or actuated externally via physical stimuli?
• What is the computational power of small robots? How much information can be stored, how fast can it be processed? As the computing power of small systems necessarily is very limited, nanorobotics means 'robotics without a brain'. Nanorobots will tend have 'embodied intelligence', in which input-output relationships between sensor and actuator functions are custom-made and hardwired rather than freely programmable.
• Can one realize collective behaviors of large numbers of nanorobots—and how can one 'program' such systems?

Ultimately, the development of micro- and nanorobotics will lead to advancements in many application areas:

• Nanomedical robots operating in living organisms will be able to deliver potent drugs at the right spot, and in a context-dependent manner. They can also act as sentinels, permanently monitoring, recording, and reporting the presence of disease indicators.
• Small robots will extend the capabilities and enhance the sensory spectrum of larger robots. They will constitute the molecular interface of hybrid robotic systems, which enables bidirectional communication with biological systems. Such systems can be integrated (e.g., with mobile robotic units that perform environmental sensing and monitoring).
• Robotic components extend the functionality of 'smart' materials and surfaces. Embedding robotic functions into materials will enhance their ability to adjust to their environment and change shape and mechanical properties in response to environmental cues. Conceivably, such materials will have something akin to a metabolism that supports these functions and enables continuous operation. Surfaces coated with nanorobotic components will have enhanced sensor properties; they may be able to actively transport matter, take up or release molecules to the surroundings, and change their structure and physicochemical properties.
• Not the least, work on bioinspired small robotic systems will elucidate physical limits of sensors and actuators, and will clarify what it takes to display intelligent (or seemingly intelligent) behavior; it will also result in methodologies that allow programming the behavior of dynamical, out-of-equilibrium systems. Ultimately, the realization of bioinspired robots may also contribute to a better understanding of complex biological phenomena and behaviors.

The provision of appropriate healthcare is unquestionably a major worldwide societal challenge that impacts all nations and peoples. The growth in medical robots driven by advances in technologies such as actuation, sensors, control theory, materials, AI, computation, medical imaging, and of course supported by increased doctor/patient acceptance may be a solution to a number of these convergent problems that include:

• Aging society and the increasing burden of dementia. This has been recognized in developed countries for many years, but is now becoming increasingly common globally. According to the OECD around 2% of the global population is currently over 80, and this is expected to reach 4% by 2050. In Europe, people over 80 already represent around 5% of the population and are expected to reach 11% by 2050 (OECD 2020).
• Increased healthcare spending and costs. According to the OECD in almost every country spending on healthcare consistently outstrips GDP growth. It has risen from 6% in 2010 to a predicted 9% of GDP by in 2030 and will continue to increase to 14% by 2060 (https://www.oecd.org/health/healthcarecostsunsustainableinadvancedeconomieswithoutreform.htm) (Moses et al 2013). The per capita health spending across OECD countries grew in real terms, by an average of 4.1% annually over the 10-year period between 1997 and 2007. By comparison, average economic growth over this period was 2.6%, resulting in an increasing share of the economy being devoted to health in most countries (OECD 2009).
• Shortage of workers in medical and social care professions. A relative decrease in the proportion of healthcare workers is increasing the demands on the active workforce, which is itself ageing in many countries. Data from the World Health Organization (WHO) indicates that the average age of nurses is over 40 in several European countries. Furthermore, the WHO states that the global health worker shortfall is already over 4 million (WHO 2008).
• The rapidly growing population and the need to provide basic medical needs in developing countries.
• Changing family structures. Family sizes are much smaller than in the past and the percentage of elderly people living alone is rapidly increasing.
• Increased acceptance of technology (Abou Allaban et al 2020a)—Although there is among many groups (particularly elderly users) an ambivalence about the use of robot technology and a preference for human touch, there is little hostility, and among younger people who are more familiar with computers, AI, smart technology, and advanced communications, there is a growing acceptance of the benefits of robotic and related technologies (Wu et al 2014). This increasing acceptance, and indeed in some instance reliance, on robots has been accelerated by COVID-19 (Zemmar et al 2020).

Against all of these demands and counter demands there is a widespread belief that medical and healthcare robotics is essential to transform all aspects of medicine—from surgical intervention to targeted therapy, rehabilitation, and hospital automation (figure 5.10).

Zoom In Zoom Out Reset image size

As the demands on health systems grow, it is perhaps inevitable that we should turn to technology and particularly robotics, both to provide the extra capacity/productivity that will be needed by aging and rapidly increasing populations, and also to continue to enhance and improve the quality of life provided by the healthcare systems. Although robots and robotic systems represent a significant investment cost, experience in others sectors, such as manufacturing, has demonstrated that the use of robotic technology can also offer significant savings and increases in efficiency/productivity, while also contributing to the establishment of high-quality, sustainable, and affordable healthcare systems. Important application domains that could benefit include medical training, rehabilitation, prosthetics, surgery, diagnosis, and physical and social assistance to disabled and elderly people (Stahl et al 2016, Wang et al 2021).

Surgical robotics

Robotic surgery involves using a computer-controlled motorised manipulator/arm that has small instruments attached to this assembly. Using artificial sensing this arm can be programmed to move and position tools to carry out surgical tasks. The surgeon may or may not play a direct role during the procedure. The history of surgical robotics dates back almost 40 years and arose from a convergence of several key advances in technology (Satava 2002).

Minimally invasive surgery (MIS): Driven by the potential to create smaller incisions, with lower risks of infection, reduced pain, less blood loss, shorter hospital stays, faster recovery, and better cosmesis, the first laparoscopic cholecystectomy was performed in the mid-1980s (Antoniou et al 2015). The benefits of this approach over conventional open surgery quickly became apparent to surgeons, patients, and healthcare providers. However, although the potential benefits were clear there were also several technical and human factors problems. These included:

• Poor visual access and depth perception, due to the quality of the 2D cameras and displays.
• Difficult hand-eye coordination due to the fulcrum effect of using long instruments inserted through a cannula. This produced motion reversals, and a scaled and limited range of motion that was dependent on the insertion depth of the tooling.
• Little or no haptic feedback and a reduced number of degrees of freedom and dexterity.
• Camera instability and loss of spatial awareness within the body cavity (the which way is up problem).
• Increased transmission of physiologic tremors from the surgeon through the long rigid instruments.

As a consequence, when performing MIS a surgeon must master a new and different set of technical and surgical skills compared to performing a conventional procedure.

VR, telepresence, and telesurgery: Around the time of the first developments in MIS, NASA was studying options for providing medical care to astronauts. Their research teams were particularly interested in using emerging concepts in virtual reality (VR), haptics, and telepresence. Teams within the US Army (Satava 2002) became aware of this work and were interested in the possibility of decreasing battlefield mortality by bringing the surgeon and operating theatre closer to the wounded soldier (figure 5.11).

Zoom In Zoom Out Reset image size

Robots in surgery: Although the first MIS was only performed in 1987, the history of surgical robotics does in fact slightly pre-date this, with Kwoh et al performing neurosurgical biopsies in 1985 (Kwoh et al 1988), and in 1988 Davies et al performed a transurethral resection of the prostate (Davies 2000). While these and other surgical robots were being developed, clinicians working with the various robotics teams realised that surgical robots and concepts in telepresence/telesurgery had the potential to overcome limitations inherent in MIS by:

• Using software to eliminate the fulcrum effect and restore proper hand-eye coordination. At the same time the software made movement and force scaling possible so that large movements or grasp forces at the surgeon's console could be transformed into micro motions and delicate actions inside the patient.
• Increasing dexterity using instruments with flexible wrists (designed to at least partially mimic human wrist action) to give increased degrees of freedom which greatly improves the ability to manipulate tissues.
• Using software filtering to compensate for any surgeon induced tremor, making increasingly more delicate operations possible.
• Improving surgeon comfort by designing dedicated ergonomic consoles/workstations that eliminate the need to twist, turn, or maintain awkward positions for extended periods.

Computer-assisted surgery (Buettner et al 2020)

Tele and remote surgery (Choi et al 2018)

Micro- and nanomedical robots (Nelson et al 2010, Li et al 2017, Soto et al 2020)

Tele-healthcare involves the remote (tens of km to thousands of km) connection of different subjects involved in the healthcare process. This includes the patients, the clinicians, and others. Tele-health systems allow different kind of interactions: (i) clinician to clinician, (ii) clinician to patient, and (iii) patient to mobile health technology. Each interaction has a different purpose and requirements.

A common scenario where tele-health can provide an important benefit is when the patient or medic is not able to travel to the clinical setup, due to distance limitations, cost, poor transportation links, or other external factors. The COVID-19 pandemic has made this scenario even more real (Chang et al 2020, Leochico 2020, Prvu Bettger 2020), yet through the technology and application of tele-medicine, medical treatment, assessment, monitoring, or rehabilitation can continue to be provided without the need to come in person to the clinic. This offers important advantages in terms of infection prevention.

A second application of tele-health is continuous monitoring. Classical tele-health approaches make use of simple technologies such as phones, email, or video-chats, but the latest IoT (Internet of Things) technologies allow continuous monitoring and feedback of many different medical data and processes (Dabiri et al 2009), This makes use of developments in ubiquitous computing, cloud storage, and intuitive human-machine interaction. Tele-Health is also exploring the emerging possibility of using the sensory capabilities of wearable devices that offer a unique opportunity to enhance the monitoring and intervention abilities of medical staff and relatives (De Marchi et al 2021). This data can be analysed and sent to nurses, doctors, and public health agencies for monitoring purposes. In addition to providing a greater range of data, wearables can also reduce the demand for staff to make manual observations.

Wearable technologies for health monitoring (Best 2021)

5G technology and tele-health (Acemoglu et al 2020b)

Tele-diagnosis and monitoring (Ding et al 2020)

The elderly population is projected to quadruple between 2000 and 2050, with many experiencing mobility impairments due to physiological muscular decay or associated health conditions, such as stroke, which shows an increasing rate of incidence with the age of the subject. In its '2030 Agenda for Sustainable Development' the UN studied future housing, healthcare, employment, and social protection needs (DoE 2017), and identified mobility as being of paramount importance to quality of life, social inclusion, and independence (Richardson et al 2015). While walking sticks, wheelchairs, or walking frames can provide reasonable assistance, many elderly/disabled still require additional assistance when walking (Charron et al 1995).

Stroke is the leading cause of disability in industrialised countries. Fortunately, over 65% of patients survive, but the majority have residual disabilities, with up to a third having severe disabilities. Hemiplegia, the most common impairment resulting from stroke, leaves the survivor with a stronger unimpaired side and a weaker impaired one (hemiparesis). In addition to stroke, traumatic injuries as well as conditions such as muscular dystrophy, arthritis, and regional pain syndromes also add to the major causes of disability and functional dependence. Deficits in motor control and coordination synergy patterns, spasticity, and pain are some of the most common symptoms of these conditions (Parker et al 1986).

Evidence has shown that intensive and repetitive physiotherapy can have substantial benefits (Carr et al 1987) including regaining motor control and muscle strength as well as in restoring/retaining the joints' range of motion. Despite the benefits of intensive physiotherapy, the associated disabilities are seldom considered life-threatening; therefore, they rate relatively low on the priority list for urgent medical assistance. In addition, manipulative physiotherapy is labour-intensive and therefore fatiguing for the therapists as well as the patient; it requires high levels of one-to-one attention in an environment with an international shortage of physiotherapists, and patients must receive individualised treatment.

Against this background, wearable technologies, robots, exoskeletons, and power- assistive techniques seem to offer a promising solution and are increasingly viewed as a potential replacement for the physical labour. This will leave therapists with greater time to develop treatment plans and enhance therapeutic outcomes.

Rehabilitation robots

Gait rehabilitation robots

Exoskeletons for rehabilitation (Pons 2008)

Social robots are often characterised as providing support or training to a person in a caring/social interaction scenario. This may be interpreted by external observers as appearing social. They are physically embodied, often in human form, although animal- and cartoon-like appearances are not uncommon. Although many social robots emulate human form, features, behaviours, and expressions, care must be taken to avoid imitating human appearance or motion too closely to avoid falling into the Uncanny Valley (Pandey et al 2018), where behaviours and appearances can become intimidating or disturbing. Social robots have some level of autonomy and can directly interact, communicate, and cooperate with people following social accepted behaviours and rules. They do not typically perform a mechanical task, although they may be capable of movement including on occasion locomotion.

For healthcare-based scenarios many see social robots as playing a part in addressing mental health issues, with the robots acting as social partners in applications such as robot companions, robots as educators for children or children/adults on the autistic spectrum (Cao et al 2019, Stower 2019), robots as assistants/companions for the elderly/disabled, and those studying affect, personality, and adaptation (Čaić et al 2019, Dawe et al 2019, Johanson et al 2020). The development of social robots can also create a bidirectional paradigm with the human learning from the robot and the robot learning from the human. This has seen particular relevance in cognitive neuroscience.

A early example of a social robot developed to support developmental psychology, cognitive neuroscience, and embodied cognition was the iCub humanoid. iCub was a child- sized robot designed to manipulate its surroundings, imitate its human partners, and communicate with them (Tsagarakis et al 2007). Over the past 20 years iCub and social robots in general have investigated human social concepts such as decision making, intent, trust, attention, perception attachment empathy acceptance, and disclosure.

Although there is a substantial body of research in social robotics with some very positive progress, there is nonetheless a substantial gap between the capabilities of social robots and the expectations of the general public, with potential users expecting s social robot to be 'like a friend' but being disappointed when the robot is not able to go beyond the smart-speaker like single-turn structure of conversation. Nonetheless the ongoing research continues to make important progress and this will only increase, resulting in robots that are more 'social' and personal over time (de Graaf et al 2015).

Service robots have expanded rapidly into many areas in the past 20 years and are increasingly common in industrial and even home settings. The healthcare sector has also been seen as an area of potential growth, although pre-COVID-19 many of these opportunities were seen as potential rather than realised goals. This was perhaps due to the challenges associated with the very nature of the services needed in healthcare. However, there is now no doubt that the COVID-19 pandemic has created an environment that has accelerated deployment and use of service robots within hospitals and across the wider healthcare system. Within this context robots are now being used either semi- or fully autonomously to provide services such as effective cleaning in hospitals/care homes and logistics of patients and supplies. This has benefits for patients, healthcare workers, and healthcare systems.

Infection control robots

Robots for healthcare logistics

Healthcare is a major societal challenge with daunting forecasts for the future, especially given the global population aging trend, the increasing number of patients, the decreasing proportion of healthcare workers, and the increasing costs of care. Fortunately, technological progress and the rise of healthcare robotics offer hope for the establishment of sustainable, affordable, and high-quality healthcare systems.

Robotics can offer significant contributions to progress in disease detection, diagnosis, and treatment at early stages; patient care; earlier more intense personalised rehabilitation; remote treatment; and the training of medical personnel, improving the skill levels of trainees and lengthening the effective career of experienced surgeons/clinicians. All of this can be achieved while enhancing the safety, precision, quality, and efficiency with which care is provided. But these opportunities do not come without challenges and barriers. There needs to be a recognition that while individual technologies can and will bring important but still incremental levels of improvement, major impacts are expected to arise from their full integration into a complete robotic healthcare system.

As noted throughout this chapter the possible areas of application of robotic technology are vast and diverse and therefore identifying generic issues and barriers to implementation is not easy. However, it is clear that there are challenges that encompass economic, social/societal, clinical, and research-related domains.

Technical

Acceptance of healthcare robotics

Healthcare robotics and regulatory approval

Jobs and healthcare robotics

Conclusions

Light for Health—what is the meaning of this title? Before we discuss this in detail, we want to introduce another term which is closely connected, namely Biophotonics. Biophotonics derives from two Greek words, namely, βιoσ life, and ϕωσ, light. Correspondingly, under the term Biophotonics we understand the application of optical technologies or methods like microscopy to solve problems in medicine and life sciences. In this sense, the use of light for health applications is nothing new, and has already established for a very long time. One cannot imagine health-related science without light, all the more, because one may extend the definition of light beyond the invisible parts of the electromagnetic spectrum, namely, into the infrared spectral range on the one side and down to the ultraviolet, and even reaching x-rays, on the other side.

Light may be used to diagnose diseases early, if possible before the outbreak, and to prevent or treat diseases (e.g., by identifying infectious bacteria that are resistant to antibiotics). But even more, light helps to understand fundamental processes in cells (e.g., by using conventional or high-resolution microscopes), and thus enables us to elucidate the causes of illnesses. Thereby it can help to fight diseases, for example, by finding the right dose of a pharmaceutical substance or finding the right drug at all. In addition, it can also be employed to track and destroy cancerous cells and tissue, and all that with minimal side effects when compared to systemic drug therapy (figure 5.12).

Zoom In Zoom Out Reset image size

What is it that renders light unique? Light is able to measure touch-free. In a strict sense, this is not completely accurate—light always changes matter, even if it is not absorbed, but usually such interactions are transient and reversible, so that light does not disturb processes in living matter it illuminates. Light interactions are instant. Instead of waiting a couple of days (e.g., for a hemogram) a result can be gained much faster within hours or, in some cases, in real time. This means it would be possible for a medical doctor to find the cause of a disease immediately and react instantly. In addition, it is the ideal tool to illuminate the world of microbes and cells, down to even sub-cellular structures.

The goals of using light-based technologies and methods for health applications are manifold. First, light-based technologies can solve the urgent problems our civilization is facing be it either strongly aging societies like (e.g., in Europe, North America, or China) or poorer societies such as in parts of Africa. This might be surprising at first glance. However, the focus is not on high-end instrumentation only, but also on portable and wearable instruments that sample data more often or even continuously. These data can be analyzed in rural areas or at the point-of-care or be remotely collected and evaluated. Such technologies can help to reduce costs and enable medical diagnosis and treatments without a medical doctor present. In any case, the direct benefit for the patient is paramount. In addition to this benefit, health-based technologies help to secure existing jobs and to create new ones. And, last but not least, light-based techniques and technologies allow us to satisfy our curiosity to understand life processes at the cellular level, to understand the cause of diseases for the ultimate goal to enable the early diagnosis and treatment of diseases.

Since there are several possibilities to subdivide the vast field of biophotonics and because it is not possible to even cover the most important methods and techniques (this would also go far beyond the purpose of this contribution), we selected examples of what we consider important technological achievements of light for health and offer an outlook of what may be possible in the future. In doing so we focus on examples of photonics approaches which are already used in clinical routines for therapy and diagnosis. Furthermore, we use the example of optical coherence tomography (OCT) to show the challenges one faces when introducing a new method into the clinical workflow. We also present an example of a very promising point-of-care approach that is close to being ready for clinical use. The latter highlights the potential of photonic approaches to address currently unmet medical needs. Finally, we close this subchapter by giving an outlook on future perspectives. However, we start with a brief introduction into the history of microscopy because modern microscopy, or the achievements in light microscopy, can be seen as one of the drivers of Biophotonics, and many biophotonic approaches involve microscopes or innovative microscopic/imaging concepts.

Understanding the world in ever smaller dimensions has spurred scientists to develop ever better optical instruments. Our current view of the microscopic world is based to a large extent on discoveries made with the help of the microscope (figure 5.13). This applies to both the animate and inanimate part of nature. Over time, the light microscope has become an outstanding instrument for medicine and the natural sciences. The time of the invention of the microscope cannot be clearly pinpointed. For more than two centuries, starting from the beginning of the 17th century, scientists and craftsmen endeavored to develop the device into the instrument we all know today.

Zoom In Zoom Out Reset image size

While in antique times the use of lenses for the cauterization of wounds is documented, it looks like as if its employment for magnification purposes cannot be attested. After the fall of the Roman empire, Arab scholars took charge of increasing the knowledge in the Middle Ages. As a first milestone, at around 1000 A.D., the invention of a reading stone took place, most likely by Ibn Al Haitham.

It is not clear who first invented the microscope. But for clarity, let's first define the term. If we accept any magnifying instrument as a microscope, then also devices with only one lens fall into this category. A stricter definition of the term requires that a microscope consists of at least two lenses, an objective lens to magnify the object and an eyepiece lens to further enlarge this magnified image. The idea of employing two lenses most likely originated in the beginning of the 17th century, but it is not clear when the first device was actually built. Instruments with two lenses were originally of inferior quality compared to instruments using only one lens, when it comes to high-power magnification. First of all, one problem was the fabrication of glasses without streaks and other defects which increased if a device with two lenses was to be constructed. A further challenge was chromatic aberration (figure 5.14). Since the refraction of light is increasing with decreasing wavelengths (dispersion), a collecting lens will have the focal spots of blue light closer to the glass than for red light. This was a common problem in early microscopes and telescopes until compound optics using differently refracting glasses were engineered. This was a big hurdle for Robert Hooke (1635–1703), who was the first to publish an extensive scientific work about microscopy called 'Micrographia'. His instruments achieved approximately 50× magnification with an incident light setting, using a cobbler's ball for illumination.

Zoom In Zoom Out Reset image size

While Isaac Newton (1643–1727) had erroneously claimed in 1666 that the problem of achromatism could not be solved, it was Antoni van Leeuwenhoek (1632–1723) who focused his developments on instruments with only one lens, with which he achieved magnifications of up to 260×. One of Leeuwenhoek's secrets was to fabricate lenses virtually free from glass defects, allowing him to reach such high magnifications that he was the first to publish sketches of living bacteria and spermatozoa. While van Leeuwenhoek did not use today's definition of what he documented, this still can be seen as the beginning of modern life sciences research.

Despite this success, the research for better microscopes lacked systematic efforts, since too many disciplines from glass fabrication to physics and engineering were involved. This changed dramatically thanks to Carl Zeiss (1816–1888), an engineer who started a collaboration with the physicist Ernst Abbe (1840–1905) in Jena, Germany. By intense theoretical studies and systematic practical investigation of the optical properties of lenses Abbe formulated the theory of (microscopic) image formation and revolutionized the entire field. Consequently, Abbe addressed optical aberrations in a holistic manner, asking for optical glass with new refractive indices and dispersion. Therefore, Zeiss and Abbe motivated the chemist Otto Schott (1851–1935) to develop optical glasses in Jena. While achromatic lenses were introduced in the 18th century by merging the images of two wavelengths (red and blue) into one plane, Abbe created the apochromatic lens by using the glasses of Schott. Here, three colors of the spectrum (red, green, and blue) are brought into focus in the same plane, while the spherical aberration is corrected for two wavelengths as well. In 1873 Abbe defined the resolution limit in wide-field microscopy as the wavelength of the light used divided by the sum of the numerical aperture of illumination and observation.

With the collaboration of Zeiss, Abbe and Schott a new age dawned, allowing the reliable fabrication of a comparably large number of high-quality optics. This started the boom of the use of the microscope in medicine and the life sciences. The differences in morphology (e.g., of a bacterium) are optically defined by differences in the refractive index only. These differences are, however, usually comparably small, which renders a lot of features hard to identify and observe. In the 19th century this issue was already solved by chemical staining of the samples. Until today, for bacteria it is common practice to color the samples by crystal violet. This staining technique is called Gram and divides bacteria into two groups: Gram-positive bacteria that become purple by Gram staining, and Gram-negative bacteria that appear pink afterward. Another example is Haematoxylin and eosin (H&E) staining, which is a standard staining technique in pathology.

In order to avoid the chemical treatment of the samples and to observe living cells phase contrast microscopy was invented in 1932 by Frits Zernike (1888–1966) and commercialized in 1941. The Nobel Prize in Physics was awarded to him for this in 1953. A few years later Georges Nomarski (1919–97) invented differential interference microscopy, a method visualizing the local differences of the refractive index of a sample.

The contrast of these methods is based on differences of the refractive index, ranging from roughly 1.35–1.5, depending on the water content, and causing optical inhomogeneities. However, these inhomogeneities often arise from differences in the chemical composition. A different kind of contrast is used by methods that are chemically specific. The most well-known property in relation to this is fluorescence. Here, the sample is illuminated by light with a short wavelength. Chemicals called fluorophores are exited by this light and re-emit the absorbed energy in a lossy process at defined higher wavelengths. This emitted light is then easily detected. The most famous fluorophore is the green fluorescing protein (GFP). It is naturally expressed in the jellyfish Aequorea Victoria as described by Osamu Shimomura in 1961, but can be prepared also outside the Jellyfish, which was accomplished first by Douglas Prasher. Martin Chalfie and Roger Tsien managed to express the cloned GFP in bacteria in the 1990s for which they received the Nobel Prize in Chemistry together with Osamu Shimomura in 2008.

Further chemical contrast methods are based on the ability of molecules to vibrate. These vibrations can be excited either by visible light (strictly speaking, molecules vibrate also without being excited, and the excitation just increases the amplitude of the vibration), where a small part of the light interacts with the molecules in this way and has a lower frequency afterwards (Raman spectroscopy), or by infrared light the frequency of which is the same as the vibrations (infrared spectroscopy). The spectral signatures allow on the one hand an identification of the molecules as a spectrum can be seen as a fingerprint. On the other hand, changes of the chemical structure can be followed. More importantly, these techniques do not require labeling as in the case of fluorescence, which avoids an alteration of the molecule and a disturbance of cell processes.

To obtain as much information about morphology, structure, and chemical content of a sample as possible, it is of advantage to combine a number of different techniques or modalities within one instrument. Such devices allow multimodal microscopy and imaging and are of particular use for medical applications (e.g., in pathology) where they can complement the classical light microscope or serve as tools for surgical procedures, where they can reveal the borders of cancer and healthy tissue.

In recent years a fundamental limit has been circumvented, which we introduced already, namely the Abbe limit. A resolution limit of about 200 nm in wide-field microscopy often does not allow to identify relevant structures (e.g., to resolve distinct distributions of actin inside dendrites and spines in the brain). One prominent method to go beyond the Abbe limit is called STED, which stands for stimulated emission depletion in confocal microscopy. Put in simple terms, by this technique the fluorescence is suppressed except in the center of the focal spot, which usually leads to a resolution of 30–80 nm (figure 5.15), but values below 4 nm have been reported in literature. For using a stochastic method with significantly less phototoxicity than STED, stochastic optical reconstruction microscopy (STORM) was introduced. For 'the development of super-resolved fluorescence microscopy' Stefan Hell, Eric Betzig, and William E Moerner were awarded the Nobel Prize for Chemistry in 2014. In the meantime, a cornucopia of different (scanning) techniques became available like photo-activated localization microscopy (PALM), structured illumination microscopy (SIM), or total internal reflection fluorescence microscopy (TIRFM), the latter making use of the evanescent field of light.

Zoom In Zoom Out Reset image size

The development of the microscope is still not at its end and further improvements can be expected to allow 3D imaging with high resolution from the components of cells up to whole organ and body imaging in video rate and beyond. It will still improve and with it its value for medicine and the life sciences.

Photodynamic therapy (PDT) involves light and a photosensitizing chemical substance used in conjunction with molecular oxygen to induce phototoxicity and elicit cell death in diseased tissue. It is recognized as a treatment strategy that is both minimally invasive and minimally toxic to healthy tissue. PDT's advantages lessen the need for delicate surgery, lengthy recuperation, and formation of scar tissue and disfigurement. PDT is applied to treat a wide range of medical conditions in skin (e.g., acne, psoriasis), macular degeneration in eye retina, atherosclerosis, and malignant cancers (e.g., skin, bladder, lung, breast, head and neck, gastrointestinal), and has also shown some efficacy in anti-viral and anti-microbial treatments.

As depicted in figure 5.16, the workflow of PDT applications involves three components and is a multi-stage process. First, a photosensitizer (PS) with negligible dark toxicity is administered, either systemically or topically, in the absence of light. Second, light activates the PS when a sufficient amount appears in the diseased tissue. The wavelength of the light source needs to be appropriate for exciting the photosensitizer. Third, the PS produces radicals and/or reactive oxygen species (ROS) that are highly cytotoxic and kills the target cells. The light dose is controlled to supply enough energy for stimulation, but not enough to damage neighboring healthy tissue.

Zoom In Zoom Out Reset image size

The key characteristic of a PS is the ability to preferentially accumulate in diseased tissue and induce a desired biological effect via generation of ROS. Among the specific criteria are also strong absorption with a high-extinction coefficient in the red/near infrared region of the electromagnetic spectrum (600–850 nm) allowing deeper tissue penetration. Whereas some PDT protocols required rapid clearance of PS within 1–2 days to minimize patient photosensitivity, prolonged production of photoreaction was reported for up to 20 days post-administration. Tailored PSs are key to the development of PDT. They are categorized as porphyrins, chlorins, and dyes. The major difference between PSs is the parts of the cells they target. 5-amino-levulinic acid (5-ALA) as member of the porphyrin family localizes in the mitochondria, m-tetrahydroxyphenylchlorin (mTHPC) as member of the chlorin family in the nuclear envelope, and the dye methylene blue in the lysosomes. PSs have also been attached to polyethylene glycol (PEG) copolymers that have a hydrodynamic size of 30–40 nm and provide combinatorial phototherapy with photothermal and photodynamic therapeutic mechanisms. To overcome the poor solubility of many PSs in aqueous media, particularly at physiological pH, alternate delivery strategies are used in new generations such as in liposomes and nanoparticles. Antibody-directed PSs have been developed to further enhance targetability.

Illumination systems are another key to the application of PDT. To achieve selected target cell destruction, while protecting normal tissues, the PS can be applied locally to the target area and globally illuminated (e.g., by exposure to daylight) or an external broadband (halogen) or narrow band (halogen with filters, laser, LED) light source. Local illumination offers another way for selected target cell destruction. For internal tissues and cancers, intravenously administered PSs can be illuminated using endoscopes and fiberoptic catheters.

Augmented reality (AR) is an interactive experience of the real-world environment where the objects that reside in the real world are enhanced by computer-generated perceptual information, including photonic modalities. AR can be defined as a system that incorporates three basic features: a combination of real and virtual worlds, real-time interaction, and accurate 3D registration of virtual and real objects. Whereas first AR experiences were introduced in entertainment and gaming business, subsequent AR applications have spanned commercial industries such as education, communications, and medicine. A potential medical AR application is to project photonic-generated data about pathological tissue margins into the surgical field of view which assists the surgeon during tumor resection. Neuronavigation is a related technique. Here, the advent of modern neuro-imaging technologies such as computed tomography and magnetic resonance imaging along with the capabilities of digitalization enabled the real-time quantitative spatial fusion of images of the patient's brain. The purpose of this technique was to guide the surgeon's instrument or probe to a selected target during neurosurgery. However, the accuracy of pre-operative recorded images was limited due to the shift of soft brain tissue, in particular after removal of larger tumor masses.

Hardware components needed for AR are input devices, sensors, a processor, and a display. Modern mobile computing devices like smartphones and tablet computers contain these elements which often include a camera and microelectromechanical motion tracking sensors such as an accelerometer, GPS, and solid-state compass making them suitable AR platforms. Input device techniques include speech recognition systems that translate user's spoken words into computer instructions, and gesture recognition systems that interpret user's body movements by visual detection or from sensors embedded in a peripheral device such as a pointer, glove, or other body wear. Various display technologies are used in AR including optical projection systems, monitors, handheld devices, and systems that are worn on the human body such as head-mounted displays and eyeglasses; even contact lenses and virtual retinal displays are in development. AR applications often rely on computationally intensive computer vision algorithms. To compensate for the lack of computing power, offloading data processing via networks to a distant machine is often desired.

First research towards AR with biophotonic tools has been reported for fiber probe-coupled autofluorescence intensity and lifetime which enables label-free and real-time identification of cancerous tissue. Figure 5.17 illustrates how the information acquired by a probe measurement technique is transformed into a two-dimensional image resulting in detailed visualization of the fluorescence-based contrast. For each location it takes 40 ms to acquire, process/analyze, and display the AR parameters on the surgeon's console.

Zoom In Zoom Out Reset image size

Such an AR approach is particularly interesting during robotic surgery which can provide high surgical precision and improve recovery time for the patient and is becoming a preferred means for numerous cancer and disease treatment. Here, due to the loss of tactile feedback, the assessment of tumor margins is based only on visual inspection which is neither significantly sensitive nor specific. The presented augmented surgery can serve as a framework for the clinical translation of other point-scanning label-free optical techniques such as elastic scattering spectroscopy, diffuse reflectance spectroscopy, or Raman spectroscopy.

Optical coherence tomography (OCT) provides structural information of specimens in depth based on the change in the refractive index between microstructures. From a layman's perspective it is akin to ultrasound imaging but using light waves instead of mechanical waves. And while in ultrasound it is possible to electronically measure the time of flight between the excitation and the reflection, to the speed of light a low-coherence interferometry approach is used. The signal is generated through light scattering in structured samples with transition zones of chemically different composed tissue locations, resulting in changes in the refractive index and imaging contrast.

To record a signal, the specimen is illuminated with a broadband illumination source with the central wavelength typically located in the region between 400 and 1500 nm. The bandwidth of the illumination spectrum is proportional to the coherence length, and as such, related to the axial resolution (i.e., shorter central wavelength with broader bandwidth), results in higher axial resolution, while longer central wavelength with narrower bandwidth results in lower axial resolution. In either case the light is considered to have a low coherence, meaning if an interferometric measurement is performed (e.g., using a Michelson interferometer) only photons that travel the same distance will interfere, resulting in a measurable signal. The photons will be ballistically and elastically scattered by microstructures in the sample and detected in the same wavelength range. Typically used sources are broadband lasers, superluminescent diodes (SLED), or halogen lamps. The excitation light is guided into an interferometric detection scheme with a fixed reference arm and, in the case of time-domain OCT, a moving sample arm. Based on the interaction with the sample and the relative length-difference between sample and reference arm interference pattern can be measured or not. The detector is recording the scattered photons from different depths of the tissue and based on a measurable interference a depth profile can be recorded. Depending on the central wavelength and the bandwidth a spatial resolution of typically 2–10 μm can be reached.

Today, time-domain OCT systems are rarely implemented and most work is performed using Fourier-domain OCT, where two possible implementations are available: spectral-domain and swept-source OCT. In the spectral domain a broadband excitation source is used in combination with a Michelson interferometer scheme. However, instead of a scanning arm spectrometer-based detection is used and a spectral interferogram, which contains the information about the depth profile of the sample, is used. This implementation significantly increases the acquisition speed, as no mechanical sample-arm scanning is required. The other method to acquire an interferogram is based on swept-source excitation, which uses a wavelength-tunable illumination source to generate the interferogram. Here, cheaper detectors can be used and problems, such as signal-roll-off in spectrometer-based implementations, does not occur, resulting in higher penetration depths.

OCT is a great example for the translation of a modality to clinical applications. As a matter of fact, OCT is one of the most widely used modalities for the diagnosis of eye-related disease, including glaucoma, age-related macular degeneration, diabetic retinopathy, diabetic retinopathy, and others. The human retina, which is a light-sensitive, multi-layered tissue located at the back of the eye, provides ideal conditions for the application of OCT. Due to the layered structure with very well-defined bands the retina is perfectly suited for analysis with OCT, and the application of the established method in ophthalmology and routine clinical diagnostics in eye diseases are performed on a daily basis. Thereby the retina is visualized as a three-dimensional, micrometer-resolved cross-sectional image with mm depth. Trained personnel can identify anatomical changes in the retinal layers and diagnose diseases, such as macular degeneration or glaucoma.

There are significant efforts to translate OCT to other clinical applications and different pathologies, such as intravascular imaging of atherosclerotic plaque, fibrotic diseases, and cancer diagnostics. While the premise of high spatial resolution with depth information is very intriguing there are methodical and technical challenges that need to be addressed first. For example, one of the problems is the access to the regions of interest where a measurement has to be performed. These locations are, except for skin, not directly and easily assessable with the typical systems and additional developments for in vivo endoscopy probes are required. There have readily been multiple implementations for in vivo endoscopy probes, which are typically based on four types of scanning geometries (i.e., micro-electro-mechanical-systems (MEMS), piezo-tube scanner, micro-motors, and optical rotary joints). The systems are technologically quite challenging, time-consuming, and usually manufactured individually. From the methodical point of view a significant challenge is the low penetration depth and the often heterogeneous, low-contrast data. The typically reached penetration depth of 1 mm is very sufficient to extract the relevant information about the layer structure. In many clinical settings the relevant features occur way below the depth of 1 mm, resulting in hard to interpret results. Moreover, tissue is not as well structured as the well-defined layers and the resulting image information is very homogeneous with very few features, making an analysis challenging. The homogeneity, at least at a depth of 1 mm, also appears when underlying changes in the physiological condition occur. An additional challenge is the interpretation of the data from in vivo studies and comparison to pathological reference information. Currently, in research it is standard to compare OCT data acquired from biopsies with pathological hematoxylin and eosin (H&E) staining of thin section slides and a one-to-one correlation is often hard to achieve. This correlation of the visual information from OCT data becomes even more difficult in in vivo applications, as correlation between measurement sites and a removed sample is not easily achieved and has to be addressed, for example, through combination with other optical reference modalities.

In summary, OCT is a good example of the translation of an optical modality to real applications, particularly in ophthalmology. Moreover, there are multiple very promising clinical applications where the method offers great potential. Nevertheless, the translation also has to address methodical and technological challenges. Specifically, the development of novel and high-resolution ultrasound transducers in combination with image processing approaches readily reaching resolutions of sub 50 μm, while providing depths of up to 10 mm. The next few years will be very exciting in terms of technological development and additional translation to clinical applications.

As mentioned above the first classification of bacteria was made possible by the advent of the earliest microscopes using different morphologies of the cells. Nevertheless, this information is limited, since only a few basic morphologies are known for bacteria. In a further step, bacteria with different cell-wall properties could be distinguished on the basis of Gram staining. Morphology and Gram staining is still used today in the routine identification of bacteria. However, with Robert Koch, the cultivation of bacteria became possible leading to the discovery of Anthrax and tuberculosis-causing bacteria. Koch also showed for the first time that the differentiation of bacteria (i.e., pathogens) based on biochemical tests was possible. This method is based on the fact that bacteria can metabolize different substances. Therefore, selective agars or enrichment agars are used to make different properties of the bacteria visible by means of indicator substances used during cultivation. The characteristic metabolic profile can then be used to distinguish the bacterial species. Today, the actual identification takes place with the help of automated systems such as the API (analytical profile index) system. The cultivation of bacteria in the presence of antibiotics with Etests or disk diffusion then provides information about potential (multi) resistances of pathogens.

In addition to cultivation-based methods, also nucleic acid-based techniques, such as polymerase chain reaction (PCR), are used to identify pathogens. Here, after biomass enrichment, first the DNA of the pathogens is isolated and then selected discriminative targets of this DNA are amplified in order to identify the species.

Cultivation-based tests are still the gold standard of bacteria identification; nevertheless, they require a lengthy cultivation time for biomass production. To shorten the complete identification process light-based methods are available. Using a short initial cultivation of only several hours, microcolonies (figure 5.18) can be used. Here the biomass of the microcolonies are dried as a homogeneous film on suitable sample holders and the pathogens can then be identified by means of the molecular fingerprint methods Infrared or Raman spectroscopy (see section 5.4.1.2). Applying a Raman fiber probe even getting measurements directly on an undisturbed microcolony on an agar plate is possible.

Zoom In Zoom Out Reset image size

When Raman spectroscopy is combined with a microscope, spatial resolutions on the order of bacteria size can be achieved. In figure 5.18(B) a dark field microscopic image of bacteria is presented. Using the excitation laser of the Raman microscope, single bacterial cells can consequently be measured. It is therefore possible to identify bacteria directly by analyzing the fingerprint Raman spectrum, without prior cultivation. Nevertheless, the bacteria cannot always be found as a pure sample. Therefore, to apply Raman microscopy it is necessary to isolate bacteria from their natural habitat. For pathogens this means destruction-free isolation methods need to be acquired for medical samples (e.g., blood, urine, or ascites) but also from environmental samples (e.g., water or soil). Despite the isolation time, identification based on Raman microscopy can be carried out in under two hours. After the identification of the pathogen species discrimination of sensitive and resistant isolates is also possible with this method. However, it is not possible to determine to which antibiotic the isolates are resistant or whether multi-resistant pathogens are present.

Microfluidic chips can be used to obtain further information on the resistance profile of individual pathogens. By combining Raman microscopy with different microfluidic chips, the changes of pathogens in the presence of different antibiotics can be analyzed. In this way, a resistance profile of the pathogens is obtained within approximately 90 min.

Besides the identification of species or the determination of resistances, microfluidic chips in combination with Raman microscopy can also be used for other applications. Since the Raman spectrum of a cell reflects normally the whole biochemical profile of this cell, the different contents in nucleic acids, protein, carbohydrates, and lipids may lead to an overall differentiation of phenotypes, growth states, or physiological states. Using this Raman spectroscopic information inside a microfluidic sorting chip (figure 5.18(C)) the cells can be then sorted into the different classes. The method is especially suitable for bacterial consortia which are heterogenous and/or cannot be cultivated. One application is the differentiation of cells with or without specific biomarkers such as special pigments, storage material like PHB, or secondary metabolites.

Another application uses stable isotopes which are metabolized during cell growth and can be recognized by characteristic changes inside the Raman spectra. Applying the isotope incorporation technique, it is possible to differentiate microorganisms according to, for example, metabolic activity or carbon metabolic pathways. In addition, the Raman-based cell sorting allows to add subsequent other analytical techniques such as PCR or whole genome analysis to the different sorted sample portions and to gain even more information about the sample.

Overall, the combination of Raman spectroscopy, chip-based sampling strategies as well as chemometric spectroscopic data analysis methods represents a powerful point-of-care approach comprising the entire process chain (i.e., from sampling to the final diagnostic result). By doing so, culture-free isolation and identification of pathogens, their host response, and their antibiotic resistance can be achieved. Thus, this promising approach offers the potential to overcome an unmet medical need by reducing the critical parameter 'time' to initiate a personalized lifesaving therapy as compared to the gold standard microbiology.

Big data—artificial intelligence

New light applications—smart photonic materials

Translational hurdles

Nature utilises 20 different amino acids to make proteins and so for an average sized-protein of about 300 amino acid residues there are 20³⁰⁰ = 2 × 10³⁹⁰ different possibilities of protein structures. There is not enough matter in the Universe to make one copy of each of these, which of course dwarf the number of unique proteins thought to exist in nature (less than 10¹⁰), let alone the number of proteins whose structures are solved (200 000 in the protein data bank). While not every random sequence of amino acids would create a protein to do something useful (just as not every random sequence of letters produces interesting literature) it is clear that nature has only sampled a tiny fraction of the possible useful protein structures that can be made. There is therefore a goal to design new proteins to carry out functions of nanomachines or form new materials, and specifically to design new proteins so far from nature that they cannot be thought to have evolved from existing sequences. Using our literature analogy, we can aspire to form completely new genres. This requires the long-sought ability to predict the three-dimensional fold of the protein—the way the polymer chain coils and bends on itself—from the sequence of amino acids that make up that chain.

In the year 2021, that ability was demonstrated to remarkable precision by the computer program AlphaFold 2, developed by Google labs (Jumper et al 2021). The program uses a deep neural network that was trained on the structures in the protein data bank—most of which were determined using x-ray crystallography at synchrotron radiation sources (Burley et al 2022) Are those sources now obsolete? The answer seems to be: no, and further experimental capabilities are needed. Proteins are not static objects, and as seen from the very first structure of hemoglobin, their conformations change as they carry out their functions, respond to their environment, and interact with other proteins. The macromolecular machines that produce ATP or transcribe DNA are exceedingly complex and intricate in their mechanisms, which are only currently understood at a superficial level. Many such machines are driven by chemical or voltage gradients, Brownian motion, or entropy, and it is not yet possible to predict how a such a system operates or how it responds in changing environments. Even the act of breaking such a mechanism—to inhibit the action of a virus or other pathogen—cannot be predicted from a static structure. This knowledge requires mapping out the conformational energy landscapes of proteins in full detail and under a range of environmental conditions or subjected to stimuli. That is, structures must be measured as they fluctuate or are directed to carry out reactions. As seen above, XFELs are answering this call, using intense femtosecond pulses to collect diffraction snapshots of small crystals, but more is required to expand the scope of these measurements and to extract structures and dynamics from large datasets.

Serial crystallography at XFELs must be expanded to achieve high-throughput multidimensional measurements. The kinetics of drug binding, for example, could be mapped by measuring diffraction at various times after mixing crystals with those molecules, over a range of temperatures. Currently the time needed per measurement point in this matrix is limited by the repetition rate of the x-ray source (which can potentially reach millions per second) and the matching rate of the detector. It may be feasible to probe tens of conditions per second. Since serial crystallography can be carried out by ejecting crystals across the x-ray beam in liquid jets, the experiment can be controlled and automated using microfluidic technologies. In this way, entire libraries of potential drug molecules or compounds could be studied, greatly expanding current activities of crystallographic screening. During the current COVID-19 pandemic, crystallography-based searches for compounds that inhibit certain viral proteins have taken place, measuring thousands of conditions over month-long campaigns, or several minutes per compound. This potentially could be accomplished in minutes. With automation comes the ability to control and use the analysis of resulting structures to guide measurements and to optimise searches. Development must be made to integrate production, purification, and chemical synthesis with the analytical capabilities of serial crystallography.

Today, molecular movies have been made for proteins undergoing reactions triggered by short light pulses, precisely timed to x-ray pulses. While such studies have given some general insights of the actions of proteins, these photoactive proteins represent but a small fraction of all proteins. The means is required to precisely trigger reactions in systems that are not sensitive to light. One promising method is to engineer caged compounds, which are released by a flash of light to begin the reaction. More generally, there are efforts to use advanced data processing and pattern matching techniques to precisely order a set of snapshots recorded at fluctuating times from experiments where the temporal evolution of a process is only loosely imposed. This relies upon machine learning to place those snapshots on a manifold, in a high-dimensional phase space, which must be discovered in the data itself. Recent tests have shown a remarkable ability to achieve a time resolution much better than the duration of the x-ray pulses, perhaps aided by temporal fluctuations in those pulses (Hosseinizadeh et al 2021). Such machine learning methods could be used to feed back into the generation of the temporal profiles of x-ray pulses, perhaps by modulating the electron beam in the accelerator, to optimise the ability to place events according to their sequence in time.

The crystallisation of proteins has been key to obtaining structures of macromolecules using x-rays from the earliest days. An alternative approach is to use a powerful electron microscope to image the molecules directly, but this suffers from a similar dependence of damage on dose as for x-rays. It is now possible to assemble 3D molecular structures at atomic resolution from very noisy low-exposure images, but this too requires cooling them to cryogenic temperatures. Nevertheless, cryo-electron microscopy is revolutionising structural biology by obtaining structures of molecules that cannot crystallise (Kühlbrandt 2014). And while structures may be distorted by the low temperatures, it is possible to sort the images into different structures to observe distinct conformations (such as a ribosome in different stages of constructing a polypeptide). X-ray free-electron lasers promise similar capabilities, but without the need to cool the molecules. This is being pursued by reducing the size of crystals in diffraction experiments down to the smallest they can go—to single molecules. The diffraction patterns of single molecules are certainly weak without the amplification provided by the crystalline lattice but, by using similar algorithms as those used to sort and assemble data in cryo-electron microscopy, it may be possible to build up diffraction datasets of single molecules that are streamed across the x-ray FEL beam as an aerosol or in a thin liquid jet. As is the case for crystals, it should be possible to trigger reactions and obtain structures as a function of time, but now molecules that follow different reaction pathways could be sorted and examined independently. From large ensembles of single-molecule data it will be possible to capture rare events that happen spontaneously, such as the precise moment a molecule binds.

High-power x-ray FEL pulses are needed for single-molecule diffractive imaging, which might be achieved by adapting chirp-pulse amplification schemes to the x-ray regime. Stronger diffraction signals will require even shorter pulses—at the scale of attoseconds—to ensure 'diffraction before destruction' of single molecules. The total pulse energy may not necessarily be much larger than what is generated today using a kilometer-long linear accelerator. With high enough control of electron beams with the next generation of particle accelerators it may be possible to shrink the facility to a size that can fit into a laboratory. Together with electron microscopes, large-scale x-ray facilities, and computational prediction, these will help to vastly expand the study of the machinery of life and usher in the new technical age of protein design—the amino age.

At any time, each person is surrounded by many pathogens. Our immune systems manage to recognize most of them in time and stop us from getting sick. Our immune systems protect us by producing a wide range of defense mechanisms ranging from mechanical and anatomical barriers, such as our skin, physical reactions such as scratching or crying, to the action of cells of the innate and adaptive immune systems. The innate immune system provides a first line of defense by recognizing stereotyped properties of pathogens. Cells recognize general properties of foreign, potentially harmful organisms such as molecules characteristic of bacterial membranes, amoeba, and other parasites. The adaptive immune system is more specific, and uses a large ensemble of cells carrying diverse receptors that are specific for pathogenic molecules. This group of B- and T-cells is called a repertoire. Importantly, the adaptive immune repertoire updates its composition based on the pathogens it encounters, meaning it is a constantly evolving and changing system. The innate system informs the adaptive system of newly detected threats, acts as a messenger between its cells, and the two systems work closely to control all infections.

There are about a billion different B-cells and T-cells, each coming in clones of different sizes, making up together about a trillion different cells that patrol our body. This diversity of cells protects us against pathogens, including those that did not exist when we were born, while at the same time being able to discriminate between molecules natural to our body (good) and those that are foreign (bad). This dynamic ensemble of ever-changing cells is self-organised and distributed, meaning there is no leader cell. Cells communicate between each other to control their numbers and proliferate in case of an infection through signaling molecules, but there is no top-down chain of command. For a few decades now, biological physicists have been studying the equations that describe the interactions between these cells and lead to the composition of the repertoire that protects us against pathogenic threats. These equations are both stochastic, meaning that the forces that drive the repertoire are random and can take different values at different times, but also the interactions are nonlinear, meaning that the existing cells influence the frequency of each other in complex ways. Together these elements make predicting the future frequencies of cells in the repertoire difficult. Physicists have a lot of experience in solving this class of equations and finding the parameters they are sensitive to and those that matter less. These skills are important for predicting future states of repertoires and correctly modelling these systems (Perelson and Weisbuch 1997).

Lastly, pathogen-recognizing receptors of the adaptive immune system are not hard coded in our DNA, like the innate system receptors. They are constantly generated by a highly stochastic process, meaning that each one of us has a set of cells that is unique to us. In fact, it is so unique it can be used to identify us. But random does not mean it can be anything. There are rules that this randomness obeys and physicists have managed to learn these stochastic rules thanks to which the functioning immune repertoire is generated. This uniqueness and randomness imply that we can all have different cells protecting us against the same viral challenge, and that they will protect us equally well. On the one hand, this means there are many good solutions to the same problem; on the other hand it means that we cannot just hope to engineer the exact same solution for everyone. The best protection helps our immune systems find its own solution, which is why both vaccines and immunotherapy work. In summary, this large but finite and constantly changing army of cells controls most of the pathogens we encounter, most of the time.

Where do these new mutant viral strains come from? From the point of view of the human population, they either are mutations from existing strains of human viruses, as in the case of seasonal influenza, or come from spillover of a virus from animals (called zoonosis) through adaptive mutations, as in the recent H1N1 influenza pandemic or SARS and MERS betacoronavirus outbreaks. Yet, even the strains that come from animal reservoirs come from mutations: before it mutates, the strain that infects animals is not well adapted to humans, and specific mutations allow it to infect and spread through the human population.

The rules underlying evolution are simple. Mutations (small local changes to the organism's DNA) and recombination (rearrangements of different fragments of DNA) generate diversity. For many microbes such as viruses, mutations play a dominant role. These changes are selected according to the survival and reproductive success of organisms that carry them. Mutation first appears in one organism in the population and then spreads with the rate this individual reproduces and has offspring. The first couple of reproductive steps are very perilous since even a very beneficial mutant may not manage to reproduce, simply due to bad luck. Once the individual manages to produce enough of its offspring—and we know how to calculate the chances of that—selection takes over and its final frequency in the population depends on whether it's a beneficial, deleterious, or neutral mutant. The life of an evolving mutant virus, as of any organism, is driven by these initial random events. Due to this initial randomness, even a mutation that is quite deleterious for the organism can rise to prominence and finally be present in every individual in the population.

Mutations and recombination events are very rare. Most of the time DNA gets replicated faithfully, with only occasional errors. For example, the mutation rate when making copies of viral DNA is one every 100 000 base pairs. That number seems small, but if there are 10 000 potentially mutating positions in each virus, and if there are 25 million copies of the virus (a fair estimate for the flu), millions of sites on the viral genome will mutate in each viral replication in each infected person. Importantly, the number of currently infected hosts increases the probability of a mutation happening. The more mutations that happen, the larger the probability that one of these mutations is beneficial and helps the virus infect new hosts and spread. As a result, although viral evolution is driven by rare and random events, we can see and feel its outcome on large scales.

Ultimately, the fate of every single mutation is to either take over the whole population, or disappear. However, before that happens the organism carrying this mutation can acquire a new mutation but in a different place in its genome. Then the fate of our initial mutation depends on all the other mutations in the organism. For this reason, a viral population carrying many strains with different sets of mutations may co-evolve with the hosts for a very long time, during which the fate of each mutation individually is settled.

One outcome of the evolutionary process is that a single individual and its descendants take over the population—every virus in the population is the offspring of the same ancestor in which the mutation appeared. While this seems very counterintuitive it is an example of the rich-get-richer law and a consequence that individuals with a selective advantage will have more offspring than other individuals. In many practical cases, especially at short times, mutants with similar reproductive success can appear and none of them will completely take over the population, especially if this population is large and not well mixed. This co-existence of several strains often happens in viral populations. While some strains may have a selective advantage, new, fitter mutants are likely to appear on the background of less fit strains, reshuffling the hierarchy of strains and completely changing the race. Of course, this does not mean the new mutant will take over, since it can also share the same fate. Physicists in the last two decades have been instrumental in highlighting and understanding this regime, which is formally analogous to thermodynamic systems driven out of equilibrium (Tsimring et al 1996, Desai and Fisher 2007). This work has shifted the focus of evolution from a winner-takes-all to a more complex picture of viral evolution.

Unlike other organisms, viruses need host organisms to reproduce: they do not have their own machinery to replicate their DNA, they need to 'borrow' it from the host. After they infect a host, its immune system attempts to get rid of them, so they need to move on and infect new hosts to survive. This means that one of the ways in which viral mutations are beneficial is by making it easier to infect new hosts, or even new types of hosts as it happens when a virus jumps from one species to another. Other intrinsic properties of the virus can also be increased, such as structural stability of its proteins and capsid or how fast it reproduces. Another type of beneficial mutations, called antigenic mutations, allow the virus to escape the immune response. Beneficial antigenic mutations will change viral proteins in such a way that immune systems of hosts that were previously immunized against the virus will no longer recognize them, because their T-cells, B-cells, or antibodies can no longer bind to them. This gives the virus time to increase the in-host population size, infect new hosts, and in the long run maybe even find better mutations (figure 5.21(A)). Ultimately, it guarantees the survival of the viral population. In response, individuals infected with this new escape strain will update their immune system, exerting further pressure on the virus to find new antigenic mutations, fueling a continual arms-race (figure 5.21(B)). Biophysicists in recent years have been working on both identifying regions in viral proteins that are prone to antigenic and intrinsic mutations and using stochastic evolutionary models to understand how mutations in different regions influence viral evolution (Nourmohammad et al 2016, Chakraborty and Barton 2017). This problem bridges the molecular scale, where the biophysical properties of the amino acid residues that make up viral proteins and underlie their binding properties matter, and the population scale, where a fine understanding of nonlinear stochastic equations is needed. Of course, not every detail matters, and physicists have been working on figuring out which details are important, both through theoretical analysis and quantitative experiments.

We have been focusing on viruses, but the same evolutionary dynamics and co-evolution is valid for antibiotic resistance. Antibiotics are specialised molecules that target and lead to the death of bacteria that infect their hosts. However, bacteria, like all other organisms evolve and beneficial mutations are such that make the bacteria reproduce even in the presence of antibiotics. To overcome this newly evolved resistance, host organisms are either given new types of antibiotics, or higher doses. Physicists are also actively studying this problem using similar approaches as for viral-host co-evolution and have set up fascinating experiments where they identified the mutations that lead to antibiotic resistance even at high doses (Bush et al 2011). This kind of quantification is useful for guiding the design of efficient antibiotic regimes to suppress the evolution of antibiotic resistance. Interestingly, often mutations in different places on the DNA work together to produce stronger beneficial effects. This can help us make predictions, as we will see below.

Since the main events that drive the evolution of pathogens are random, is there any hope in predicting its outcome? In the last couple of years physicists have taken the idea of predicting influenza strains seriously (Łuksza and Lässig 2014, Neher et al 2014). They have shown that by taking a probabilistic approach combined with biophysical knowledge at the molecular scale and stochastic equations of evolution, one can predict this year's dominant flu strains, and one can even predict the frequency of these strains compared to other strains. Although viral evolution is driven by random rules, this does not mean everything is equally likely to happen. There are stochastic rules that make certain outcomes more likely than others. We can explore our knowledge of molecular biophysics and stochastic processes to assign probabilities to the outcomes of viral evolution. Specifically, starting with flu strains that have been measured in February of a given year, they use stochastic evolutionary models to calculate the frequency of each strain in the autumn of that year, when the flu starts spreading in the Northern Hemisphere (and analogously in August for the southern hemisphere). These equations account for the competitive interaction of the different strains but also for the interaction with the immune system. Since these are not exactly known, different scenarios must be considered. The basic idea is similar to the one used in good portfolio design or weather prediction: we do not put all our eggs in one basket but sum over all possible outcomes, weighted by how likely they are. Taken together, the stochastic rules work well and these methods are now used by the WHO to decide which flu strains should be targeted in the flu vaccine for a given year.

Another example of quantitative random rules learned by physicists (Murugan et al 2012) describe the stochastic process that drives the huge diversity of B- and T-cell receptors and antibodies that the body can produce. These receptor proteins are encoded in the DNA. However, if we wanted to encode a billion different receptors in the DNA, the DNA would physically not fit into the cell. Instead, the DNA is edited in each B- and T-cell in a way that pre-existing gene building blocks are assembled like legos to produce hundreds or thousands of different receptor combinations. Additional random insertions and deletions of base pairs further increase diversity, resulting in the billion different receptors in a given person. Again, although the rules are random, not every outcome is equally likely. Physicists have quantified these random rules and can calculate the probability of any of us producing a concrete B- and T-cell receptor. Since the process is random, different people have different sets of billions of receptors. But knowing these probability rules allows us to predict that some receptors are more likely to be shared between different people, simply because they are more likely to be produced. These widely shared receptors and antibodies are called 'public'. In general, the machinery can produce much greater diversity than is realised in one person. Even the population of the whole world is not enough to exhaust all the possible receptors that can be generated. However, this is yet another example that random does not mean unpredictable. We can predict exactly how many receptors will be shared between different sets of people, and even which ones (Elhanati et al 2018).

In predicting future flu strains we solve one part of the prediction problem: knowing existing viral strains, can we predict their future frequencies. However, can we also predict new mutations that will appear and estimate how likely they are to survive the competition with other mutations and the challenges of the immune system? This is an important question for predicting escape mutations: mutations that are not recognised by the immune system. One approach that has been pioneered by physicists is making quantitative measurements of binding between viruses and B-cell receptors simultaneously in large libraries of immune cell mutants (Adams et al 2016). These methods have been extended to measure the binding properties of all possible single mutants in a key SARS-CoV-2 protein (the receptor binding domain of the spike protein) (Starr et al 2020). These measurements allow us to predict the ability of the virus to enter human cells, which is linked to disease severity and transmissibility, including for mutant strains that haven't emerged yet. These types of quantitative measurements allow us to explore the evolutionary trajectories of viral evolution and show which mutations close off paths for future mutants. Such knowledge allows us to propose vaccines that can guide the immune system. Physicists have shown that having such knowledge is important to design vaccine schedules that elicit protective immune cells (broadly neutralizing antibodies) that protect us against many viral mutants (Wang et al 2015).

More generally, an important challenge is to measure or predict the binding affinity of immune receptors and antibodies to their target on pathogens, which in turn determines the efficacy of these receptors to fight the disease. Since both the number of receptors and antigens are extremely large, it is impractical to measure all pairs one by one. Massively parallel methods need to be designed, and biophysics-inspired computational methods need to be developed to fill in the 'gaps' of receptor-antigen pairs that we will not be able to measure. Promising methods that combine droplet microfluidics (Gérard et al 2020) and binding assays (Zhang et al 2018) are paving the way to that goal. The idea is to encapsulate single cells in small droplets of water suspended in oil, and then manipulate these droplets in a microfluidic chip to perform biochemical reactions in each cell in a compartmentalized way. By combining phenotyping and binding assays in each droplet in this way, and by introducing unique molecular barcodes prior to sequencing, one can associate the full sequence of the receptors to its affinity for a set of antigenic targets, which can themselves be sequenced. To complement these experiments and extrapolate their results for sequences that were not directly tested, we expect machine learning methods to be useful. As they are improved and scaled up in the future, these techniques will help to build a complete specificity map between receptors and pathogens, which will serve as a basis for rational drug and vaccine design.

As we see, the immune system and viruses are constantly co-evolving: the immune systems of the world's hosts exert pressure on viruses and viruses force our immune systems to constantly update itself. Vaccines prepare our immune systems for challenges that we have not yet seen, such that if we encounter the virus we have pre-existing protection. Additionally, the immune system is flexible and protection against one virus can also protect us against similar viruses, as was originally shown by Edward Jerne using the cowpox virus to immunize people against smallpox in the 18th century. Similarly, using sequencing measurements and analysis, physicists have shown that T-cells reactive to the common cold, which is also a virus from the coronavirus family, can recognize SARS-CoV2 proteins (Minervina et al 2021). Using quantitative experiments to understand the range of this cross-reactivity is an important challenge that can help us predict and trigger immune responses. This may also allow us to extend the prediction of viral strains to longer timescales.

There are many different molecular solutions to the same challenge, yet we show that we can statistically predict the immune pressures exerted on evolving viruses. How can we reconcile this large molecular diversity with predictability? One answer is that the phenotype is more constrained and reproducible between experiments than the molecular implementation in terms of actual mutations in the DNA (Lässig et al 2017). In short, since the selection pressure acts on the phenotype, it is constrained, and thanks to that, predictable. As we showed in the example of the immune system, there are many ways to obtain the same results in terms of mutations. The same results have been observed for antibiotic resistance and E. coli evolution under the pressure of the innate immune system in mouse guts.

In summary, host-pathogen co-evolution occurs on many different scales: from the molecular, cellular, organismal to the worldwide population scale. On most of these scales we are dealing with random interactions between many elements. However, thanks to experimental and theoretical work we now have a command of these stochastic rules which allows us to make statistical predictions. Since this co-evolutionary race is constantly happening, why do we not see pandemic outbreaks all the time? We explore this question in the next section.

Imaging was the first application of x-rays, with the famous pictures of Mrs Röntgen's hand published on The New York Times on January 16, 1896. In over a century of progress, figure 5.22 shows the progress in imaging from the planar x-ray image produced with the orthovoltage tube of Röntgen to the modern CT angiography.

Zoom In Zoom Out Reset image size

Imaging has therefore changed medicine maybe more than any other technical advances. Therapies like surgery, radiotherapy, and any other medical and pharmacological intervention strongly depend on the imaging obtained with physics techniques. Imaging today follows the pathway described in figure 5.23 and uses many different techniques, as described in table 5.1.

Zoom In Zoom Out Reset image size

Different methods are used in different branches of medicine. Mammography is, for instance, specialized for the detection of breast cancer, and angiography for cardiovascular diseases analyses. Other methods are used for many different methods. Echography is for instance popular during pregnancy, but is also widely used to detect abdominal diseases. SPECT and PET are largely used both in oncology (often associated to CT) and cardiology in myocardial perfusion tests. CT and MRI are largely used in oncology, with complementary targets: while CT has a superior resolution of bones, MRI shows excellent images of the soft tissues (figure 5.24).

Zoom In Zoom Out Reset image size

When using endoradiology, image fusion is generally necessary to combine the functional and anatomical images. Figure 5.25 shows a fusion of CT and PET imaging in oncology. A whole-body PET-CT scanner has been recently built by the EXPLORER consortium, led by the University of California at Davis (figure 5.26).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The new frontiers of imaging is the improvement of both hardware and imaging parts. The main hardware challenge is the increase of the image resolution. The new ultra-high resolution CT scanners reached resolution of 150 μm, starting to touch the area of microscopy (figure 5.27).

Zoom In Zoom Out Reset image size

Similarly, for MRI there is a rush to increase the intensity of the magnetic field, which immediately reflects into increased resolution (figure 5.28).

Zoom In Zoom Out Reset image size

The cost of MR scanners largely increasing with the field intensity, ranging from less than 500 k€ for a conventional 1.5 T scanner to over 5000 k€ for a 7 T scanner. But for preclinical studies at CEA in Saclay researchers have already installed a 11.7 T MRI magnet. Whether these high magnetic fields are really needed in clinics is a big question for the coming years, where it will be defined what the 'optimal' scanner will be.

The perspectives are even more exciting for software, where we are witnessing the invasion of AI in medicine and in diagnostics. AI systems have demonstrated an extraordinary ability to capture diseases in images notoriously affected by many artifacts such as mammography. Nevertheless, the initial concerns that AI will replace radiologists have gradually vanished, considering that the answer to a single question is not sufficient to draw a solid diagnosis. However, it is clear that future radiologists will all use AI as a sharp-eyed partner to assist physicians in diagnostic procedures.

Another large impact of AI in diagnostics is in the field of radiomics (i.e., the method to extract molecular and histological features from radiographic medical images using data-characterisation algorithms) (figure 5.29). The procedure is very intriguing, because in principle it can overrule the invasive and slow biopsies replacing them with deep learning from conventional radiology. Shape, intensity, and texture are typical image features that can be analysed to extract biomarkers using radiomics.

Zoom In Zoom Out Reset image size

The field of functional imaging is also the one with the highest potential. Functional imaging in oncology generally uses the cancer metabolisms, mostly using 18FDG. However, recent studies are targeting cancer-associated fibroblasts, using fibroblast activation protein inhibitor (FAPI) labelled with ⁶⁸Ga. The shift from tumour metabolism to the tumour microenvironment may provide new, important information for appropriate treatment. Other compounds (e.g., 18F-FMISO or Cu-ATMS) target hypoxic regions in tumors that are treatment-resistant.

The other popular functional imaging method is fMRI that measures changes in blood oxygenation levels, which are correlated with the activity of a specific part of the brain. Apart from initial studies on brain anatomy, fMRI is now becoming a powerful tool in diagnostics and psychiatry. SPECT can also be used for cerebral blood flow measurements using compounds such as HMPAO that converts rapidly from the lipophilic form, which passes the blood-brain barrier (BBB), to the hydrophilic form, which is unable to pass the BBB, and is trapped in the brain (figure 5.30).

Zoom In Zoom Out Reset image size

Soon after the discovery of x-rays it became clear that radiation could have been used to destroy tumors, thus starting what we call radiotherapy. The goal of radiotherapy is very simple: to provide a dose as high as possible to the tumour, in order to sterilize it, but simultaneously to minimize the dose to the normal tissue, so that side effects are acceptable. The curves representing the tumour control probability (TCP) and the normal tissue complication probability (NTCP) as a function of the dose must stay separated (figure 5.31). The role of the physics is to enlarge their distance (therapeutic window), thus making the treatment effective and safe.

Zoom In Zoom Out Reset image size

Figure 5.32 gives a schematic view of the progress in radiotherapy in the past century focussing on the most common male tumour treated in the clinics: prostate cancer. Radiotherapy starts with orthovoltage x-rays that cross-fire the tumour target in a box. In fact, the depth–dose-distribution of x-rays is unfortunate: the entrance dose is always higher than the dose deep in the body, where the tumour is situated (figure 5.33).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Cross-firing from different angles is therefore necessary to enhance the target avoiding unacceptable toxicity in the entrance.

To make the depth–dose distribution more 'flat' it is necessary to increase the photon energy, and therefore the dose distribution improves with the introduction of ⁶⁰Co γ-rays (about 1 MeV energy) and later with linear electron accelerators (linacs), which can reach electron energies around 25 MeV, even if the current linacs are normally operated at 6 MeV. However, the cutting-edge technology is the use of protons and heavier ions, because the depth–dose distribution is favourable (figure 5.33). Charged particles deposit indeed most of their energy at the end of the range (the Bragg peak), where the tumour is situated. The exact position of the Bragg peak can be modulated by changing the accelerator energy, and the Bragg peak can be widened (spread-out-Bragg-peak, SOBP) to cover the whole tumour size (generally several cm) by superimposition of beams of different energy.

In addition to the physical advantages, slow ions in the Bragg peak are biologically more effective than fast ions in the entrance channels, whose radiobiology is more similar to x-rays. For this reason, heavy ions are particularly indicated for radioresistant, hypoxic tumors.

As a consequence of their physical and biological advantages, particle therapy is growing worldwide. According to the statistics of the Particle Therapy Co-operative group (PTCOG), in January 2021 there were 111 particle therapy centers actively treating cancer patients, but only 12 using ions heavier than protons (C-ions). Despite the favorable dose distribution shown in figure 5.32, only about 1% of prostate cancer patients are treated with particles. The main problem remains the high cost. Particle therapy centers require indeed large circular accelerators (cyclotrons or synchrotrons) and heavy gantries for beam delivery at different angles (figure 5.34). As a result, the cost of a heavy ion therapy can be as high as 200 M€, versus only 5 M€ for a conventional linac.

Zoom In Zoom Out Reset image size

In addition to external beam therapy (teletherapy) there has been strong progress in target radionuclide therapy, an internal therapy based on the idea of delivering molecules labeled with radioisotopes to the tumor. The radiopharmaceutical is comprised of a radioisotope with appropriate half-life, a linker, and a small molecule (peptide, antibody, etc) able to target cancer cells. A gold standard for targeted β-therapy is ¹⁷⁷Lu, which has been used with success in many tumors including neuroendocrine and advanced prostate cancer.

It is generally acknowledged that the number of cancer patients receiving radiotherapy (slightly more than 50%) is too low and should be increased in the coming years. While most patients will be treated with IMRT, particle therapy is at a tipping point. There are many efforts to reduce the cost of using superconducting magnets for synchrotrons with reduced footprints. Laser-driven particle accelerators are still immature for medical applications, but they bear the potential of having table-top accelerators for particle therapy, even if this goal goes probably beyond 2050. The same seems to be true for other innovative accelerators such as dielectric wall or fixed-field alternating gradient (FFAG) accelerators.

Beyond the cost, particle therapy has to meet the challenges posed by innovative methods in conventional radiotherapy, which is rapidly improving thanks to image guidance. MRI-linacs are already in use in clinics, and allow treatment of patients with full control of organ movements using MRI imaging. Particle therapy is still lagging behind in image guidance, even if the nuclear interaction of charged particle offer beam monitoring and range verification opportunities that are not possible with photons, such as secondary radiation, proton radiography, and PET.

This includes prompt γ-ray detection to assess the beam range; dedicated monitors are already commercially available. Particle radiography is another attractive possibility, both for imaging and for the possibility of reducing the uncertainty in the conversion between the Hounsfield units coming from the CT and the water-equivalent path length needed for particle treatment planning. PET has been used already for C-ion therapy and would greatly benefit if therapy could be done using radioactive ion beams (e.g., ¹¹C or ¹⁰C), positron emitters that can be produced by fragmentation of stable ¹²C. The results would be an enormous increase in the signal-to-noise ratio and the possibility of monitoring online the beam delivery in the tumour (figure 5.35). The use of radioactive beams in therapy had already been proposed in the 70s at the Lawrence Berkeley Laboratory, but has always been hampered by the low intensity of the radioactive beams. The recent intensity upgrades in the accelerators and the push for improving precision of particle therapy has raised interest in this project. CERN has a project for a cyclotron producing radioisotopes to be injected in conventional medical synchrotrons, whilst GSI, exploiting the intensity upgrade for the construction of the new FAIR accelerator, is actively measuring these beams in the framework of an EU ERC Advanced Grant.

Zoom In Zoom Out Reset image size

In targeted radionuclide therapy, the new challenge is probably the use of α-emitters. Being short range and with high biological effectiveness, α-particles are indeed ideal for effective tumour cell killing with reduced damage to the surrounding normal tissue compared to the long-range β-particles. Some of the clinical results obtained with radioisotopes such as ²¹¹At, ²¹³Bi, ²²³Ra, ²¹²Bi, ²²⁵Ac in metastatic patients have been stunning and have attracted great interest in this target therapy modality. The main challenge for this therapy is actually the radioisotope production. Availability of radioisotopes is a general problem for both diagnostic and therapy because many of them are produced in nuclear reactors that are in the process of being shut down in many countries. Even ^99mTc, the most commonly used radioisotope in the world for radiodiagnostics, is insufficient. The problem is even more grave for a-emitters: for instance, ²¹¹At is only produced at ARRONAX, the radioisotope factory in Nantes (France). Radioisotope production will certainly be a major challenge in the coming years.

Magnets have been the most important application of superconductivity, and accelerators can certainly be credited to be among the key drivers of the development of this technology.

Accelerator technologies, and thus accelerator magnets, need long R&D with programs spanning over decades, which allows investigation and R&D to be done in a coherent way and thus to be fruitful. The latest example, the LHC, is the summit of over 30 years of development of superconducting magnets (SCM). Its enormous size and its long-awaited goal, the Higgs particle, whose discovery about ten years ago at the Large Hadron Collider (LHC) at CERN (Last update in the search for the Higgs boson 2012) have been heralded worldwide, and the possible unveiling of the new world beyond the Standard Model make the LHC (LHC DESIGN Report 2004, The Large Hadron Collider 2018) the crossroad between past and future of high-energy physics (HEP). But LHC may be the crossroad also for medical sector superconducting technology. Thanks to the experience gained with superconductivity for LHC and its predecessors, now the accelerator magnets community is exploring the possibility of using superconducting accelerator magnet technology for particle irradiation-based cancer therapy.

Since the pioneering time of Ernest O Lawrence in Berkeley, where he designed, built, and put in operation the first cyclotrons in the 1930s, the demand for higher energy accelerators has driven the quest for larger size rings and higher field magnets. Not surprisingly superconductivity has become a key technology for modern accelerators and, conversely, particle accelerators have promoted superconducting (SC) magnet technology (Wilson 1999). The timeline of record accelerator energy is shown in the compilation reported in figure 5.36, where we have collected the main accelerators for nuclear and particle physics, classifying them according to the center-of-mass energy and highlighting the projects that are based on superconductivity. Many of them, the most powerful ones actually, feature SC magnets, or superconducting RF cavities, as the main technology and performance driver.

Zoom In Zoom Out Reset image size

HEP has been pursuing superconductivity for many large projects, the most recent one being the already cited LHC at CERN, and plans are being made for further stronger and larger colliders, as shown on the right side of the timeline of figure 5.36. We wish in particular to refer to the some of the monographs (Wilson 1983, Mess et al 1996, Ašner 1999) as well as to the more recent reviews on SC magnets and conductors for accelerators (Tollestrup and Todesco 2008, Rossi and Todesco 2011, Rossi and Bottura 2012, Bottura and Rossi 2015).

As far as the medical sector is concerned, superconductivity has been limited to a few types of low-energy cyclotrons or synchrocyclotrons, mostly for hadron (protons or ions) therapy. The use of particles accelerated at a few hundreds of MeV per mass unit for cancer therapy is a well-established technique (Degiovanni and Amaldi 2015). Particle therapy can provide a powerful tool for the treatment of tumours that are not curable with conventional x-rays or that are better treated by particle beams. There are 24 particle therapy centers in Europe and 105 in all the world. However, only four of them in Europe and only 12 in the world employ heavy ions (Carbon ions), the rest employing only protons, despite the fact that heavy ions, like carbon, are effective for some tumours resistant to x-rays and protons, and are superior in preserving the healthy tissue surrounding the cancerous zone. This is due to the much larger size and cost of the required infrastructure for ions than for protons (not to mention x-ray).

As a further consideration, almost all proton centers are equipped with a gantry, allowing beam delivery from multiple directions, boosting the treatment effectiveness. On the contrary, very few ion centers are equipped with a gantry. The first, and so far unique in Europe, has been the center of HIT (Heidelberg Ion Therapy center) (Haberer et al 2004) that has been in operation since 2012 and employs classical resistive magnets, which makes it have a length of 26 m and a weight of about 600 tons (rotatable part). A step forward has been carried out by the HIMAC center that from 2018 has been routinely using a more compact ion gantry based on SC magnets, designed and manufactured by Toshiba (Iwata et al 2012). This has allowed the Japanese center to reduce the size and weight to about 13 m and 300 tons.

A further reduction of size and weight, which eventually would reflect in reduction of cost, would certainly enable a wider diffusion of ion gantries, thus contributing to better coverage of the territory by ion therapy centers. A large collaboration in Europe is looking to leverage the experience gained in the HEP laboratories for collider SC magnets to try to make a bold step and design a superconducting gantry for ion less than 100 tons in weight. This would allow to reduce the direct cost of the gantry and footprint of the infrastructure. This application would be the second, more direct, beneficial spin-off from HEP accelerators into the medical sector, after the development of the superconductor for the Tevatron that was essential for enabling superconducting MRI magnets. Today MRI is a 5 billion dollars business, and thousands of the large magnets fabricated each year for large size imaging are superconducting ones, thanks to the high-quality, multistrand, high current density superconductor developed first for Tevatron.

HEP synchrotrons and especially HEP collider rings are the most challenging applications for SC magnets, as field quality, stability, and cost are very demanding and by far superior to other systems (like Fusion, MRI, high-field solenoids, etc). They are superconducting, with a few exceptions such as fast-cycling synchrotrons like the J-PARC complex; however, the fast-cycling FAIR-SIS100, for example, is superconducting as well. Magnets for superconducting linear colliders such as ILC, XFEL, and ESS are superconducting, too, despite the very low-field/gradient, because of the big gain in compactness and integration within the cryo-module (cryostat) hosting the superconducting radio-frequency cavities. There has been considerable effort in SC magnets for low-energy accelerators for nuclear physics, like cyclotrons or synchrocyclotrons, an effort that continues mainly driven today by medical applications. These SC magnets are similar in design and layout to magnets for MRI or particle detectors for momentum spectrometry, having the shape of solenoid or circular split coils. Here we will limit our consideration only to SC magnets for HEP colliders and their implication for the medical sector.

Superconducting magnets become of interest when the required field is above the iron saturation limit (i.e., approximately 2 T). In some instances, however, SC magnets are becoming a choice of preference for their compactness and their low-energy consumption also at low field, a topic that is increasingly important in the design of new accelerators.

Dipoles

The first function of a magnet is to guide and steer charges particles (i.e., to provide an adequate centripetal force to keep them in orbit in a circular accelerator or just to bend them in a transfer line; see figure 5.37, left). The adequate force can be given only by a transverse magnetic field, except for very low-energy beams, where solenoids and sometimes electrostatic deflectors are used. Transverse field means a magnetic field whose main component is perpendicular to the particle trajectory. For high-energy accelerators, the beam region is a cylinder that follows the beam path and with the smallest practical dimension, as shown in figure 5.37, right. As discussed later, the cost and technical complexity of the magnetic system are proportional to the energy stored in the magnetic field, which explains why it is important to minimize the size of the magnet bore.

Zoom In Zoom Out Reset image size

Even though static magnetic fields do not accelerate, in circular accelerators the bending (dipole) field eventually determines the final energy reach. In relativistic conditions the relation between the beam energy, E_beam in TeV, the dipole field B in T, and the radius of the beam trajectory inside the bending field R in km takes a very simple form:

$$\begin{eqnarray*}&&{E}_{\mathrm{beam}}\cong 0.3BR\end{eqnarray*}$$

Since the dipole field typically covers 2/3 of the accelerator, R is about 2/3 of the average radius of the ring. The above relation clearly shows the interest for the highest possible field for a given tunnel.

Figure 5.38 shows an artistic sketch of the LHC dipole, which is called the Twin-Dipole since it hosts two opposite dipole fields in the same magnet, to bend the two counter circulating proton beams of the colliders. There are 1232 main dipoles installed in the LHC ring.

Zoom In Zoom Out Reset image size

Quadrupoles and high-order magnets

The other key function of magnets is to provide the stability of the beam in the plane transverse to the particle motion. This is realized by the use of quadrupoles, sextupoles, octupoles, and in certain cases of higher-order harmonic magnets; in the LHC, magnets up to the dodecapole order are employed to fulfil the request of the sophisticated beam dynamics. For example, the action of a quadrupole on a particle beam is schematized in figure 5.39, left, showing the effect of the so-called strong focusing on a particle beam. In figure 5.39, right, a picture of the cross-section of one of the 400 main quadrupoles is shown.

Zoom In Zoom Out Reset image size

While the main quadrupoles are usually of similar—although reduced—size and complexity with respect to the main dipoles, sextupoles and octupoles are of smaller size, peak field, and stored energy, thus featuring less complexity than the main dipoles. It is worth noticing that the quadrupoles that are in the so-called insertion regions, just before the collision points, usually called the low-β section, are necessary for the optics manipulation to focus the beam to the smallest attainable size (i.e., the minimum β-function) at the collision point. This set of quadrupoles must have an aperture significantly larger than the lattice quadrupoles and requires an integrated field gradient much larger than the regular lattice quadrupoles. Larger aperture and higher gradient result in a peak field that could be close to that of the main dipoles. In exceptional cases, like the High Luminosity LHC project, actually the low-β insertions around the ATLAS and CMS experiment would be equipped with quadrupoles of much larger peak field than the lattice LHC dipoles.

Accelerator dipoles and quadrupoles: design and field evolution

Superconductor

There are tens of thousands of superconducting materials. However, only a few of them have characteristics that make them usable in real devices (i.e., with a critical temperature well above 4.2 K, LHe normal boiling temperature), critical fields above 10 T, critical current density above 1000 A mm⁻² at the operating temperature and field [e.g., for LHC is 1.9 K and nearly 9 T]). In addition to these characteristics, the superconductor for accelerator magnets needs: (i) 3–20 μm effective filament diameters, to maintain a good field quality during the ramp from injection to flat top; (ii) a geometry and a ductility such as to be assembled in compact cables; and (iii) to be robust enough to be wound in a coil and to withstand mechanical stresses largely exceeding 100 MPa. In figure 5.40 the engineering critical current density J_e of the few available practical superconductor is plotted versus magnetic field. J_e is defined as J_e =J_c ·ff where J_c is the critical current density of the superconducting material and ff = A_SC /A_tot is the volumetric fraction of superconductor in the wire (that contains also stabilizing copper and other passive materials, like barriers and alloys). As is known, all superconducting magnets in existing accelerators are wound with Nb–Ti superconductor, High Luminosity LHC being the first project breaking the 10 T-wall (i.e., attempting to use Nb₃Sn for collider magnets rated for 11.5 T, and MgB₂ for the powering line; Bottura et al (2012), Rossi and Bruning (2016)).

Zoom In Zoom Out Reset image size

In figure 5.41 a cross-section of the Nb–Ti/copper wires used in the LHC magnet is shown, together with the cable grouping many wires to reach at least 13 kA at 11 T, 1.9 K. The flat cable shown in figure 5.41 whose aspect ratio is ∼10 and has 20–40 wires has a slightly trapezoidal cross-section with the two large faces making a 0.5°–1.5° angle, which is called the 'Rutherford cable'.

Zoom In Zoom Out Reset image size

High critical current density

Electromagnetic design and field quality

The coil cross-section is usually a circular sector filled with Rutherford cable arranged with spacers to mimic a distribution of current J₀cosϑ (i.e., a current density that is maximum at the midplane, J₀, decreasing like cosine function when increasing azimuth, and becoming zero at the pole [vertical axis]; Rossi and Todesco 2007, 2006). This is the most efficient distribution to generate transverse field in an infinitely long cylinder. The coil aperture being 50–100 mm and the magnet length being 5–15 m, the approximation is very good, indeed!

In figure 5.42 we show the cross-section of the main dipoles for the most important hadron colliders and a figure of a real LHC main dipole cross-section. A perfect J₀cosϑ would yield a pure dipole field. A perfect J₀cos2ϑ (i.e., with maximum at 0°, 90°, etc) would yield a perfect quadrupolar field, a J₀cos3ϑ current distribution would yield a perfect sextupole field, etc. The winding cannot be perfect, of course: however, the requested relative field accuracy is of the order of 10⁻⁵. This calls for very accurate design and tight control of the tolerances. In the LHC all components have very strict tolerances from a few μm (superconductor) to 15–30 μm for the biggest mechanical components.

Zoom In Zoom Out Reset image size

In addition, the bending strength of each dipole must be equal to nearly 0.01%, because a full sector (154 dipoles in the LHC) is powered in series, This is actually one of the main constraints of the use of SC magnets in accelerators (e.g., the RF cavities in a linac may differ from each other in integrated gradient by 5% without serious drawbacks). In hadron collider the single weakest dipole determines the energy of the beam.

Protection and quench

The main issues for HEP magnets are the large current density in the conductor and the large magnetic energy stored per unit of coil mass. When a magnet undergoes a sudden irreversible transition from superconducting to normal conducting state, which is called 'quench', the magnetic stored energy is turned into heat via Joule effect. SC accelerator magnets can tolerate temperature increases up to 300 K, if well-conceived and manufactured. The hot-spot temperature T_{hot spot} can be estimated using an adiabatic heat balance, equating the joule heat produced during the discharge to the enthalpy rise of the conductor:

$$\begin{eqnarray*}&&{\int }_{{T}_{{\rm{op}}}}^{{T}_{{\rm{hot}}{\rm{spot}}}}\frac{C}{\rho }dT=\frac{1}{{A}_{{\rm{Cu}}}{A}_{{\rm{tot}}}}{\int }_{{t}_{{\rm{quench}}}}^{\infty }{I}^{2}dt\end{eqnarray*}$$

where C is the total volumetric heat capacity of the superconductor composite, and ρ is the resistivity. The speed at which the energy is dumped depends on the magnet inductance and the resistance of the circuit formed by the quenching magnet and the external circuit R = R_quench + R_ext. The characteristic time of the dump is then τ ≈ L/R, which can be made short by decreasing the magnet inductance by use of large current cables. Protection (i.e., fast dump to avoid excessive T_hotspot) is then achieved by triggering heaters embedded in the winding pack, and fired at the moment a quench is detected to spread the normal zone over the whole magnet mass. In the LHC the typical time to detect a quench is 10–20 ms and the heaters must activate the quench in 100 ms or less. A further complication in collider is due to the fact that the main magnets are powered in series, and the stored energy in a single circuit is much larger than the few megajoule energy stored in a single magnet. In the case of each of the eight circuits formed by the series of 154 LHC main dipoles, the stored magnetic energy attains the gigajoule level, which was the reason for the gravity of the above-mentioned LHC incident of 2008 (Rossi 2010).

As previously mentioned, LHC has been the summit of 30 years of development of accelerator magnets, all wound with Nb–Ti superconductor. For more than ten years a considerable amount of effort has been occuring in the US laboratories, BNL, FNAL, and LBNL, as well as in Europe at CERN, CEA (France), and—more recently—at CIEMAT (Es), INFN (It), PSI (Ch), to develop accelerator magnets wound with Nb₃Sn. As can be seen in figure 5.40, Nb₃Sn extends the attainable region from 8–9 to 15–16 T. The High Luminosity LHC project at CERN (Brüning and Rossi 2019, Rossi and Brüning 2019) has the main technological scope of design and manufacture, and operated the first set of Nb₃Sn magnets in a collider, ever. To go beyond this field level in the 20 T region use of HTS (high-temperature superconductor) is necessary (Rossi and Senatore 2021), as can be seen in figures 5.40 and 5.43, which summarizes the past and the possible future of HEP accelerator magnets.

Zoom In Zoom Out Reset image size

However, these technologies, Nb₃Sn and HTS, are not yet mature enough for societal applications, and in the next chapter we will deal with how the technology used for present colliders can spill into the medical sector.

As mentioned in the first section of this contribution, the use of the slim, tubular shape accelerator magnets can open the way to ion gantries of limited size and weight. Two points need to be specifically addressed for such a step into medical applications: (1) the need to ramp the field in coils that must be indirectly cooled (no use of liquid because gantry is rotating), with a ramp rate of 0.1–1 T s⁻¹ which is from 10 to 100 times faster than the sweeping rate of HEP colliders; (2) the fact that the relatively low magnetic rigidity of the beam, together with fields of 3 T or more, implies the necessity to build strongly curved magnets, which has never been attempted for accelerator magnets.

The recent rush for superconducting gantry

The CCT design

The CCT is a new magnet type that has been pursued at LBNL for HEP and for a superconducting gantry for proton beams (Brouwer et al 2020). The schematic of a CCT dipole, basically composed of two slanted solenoids, is shown in figure 5.44, together with a picture of the construction at CERN of 2 m-long prototype (Kirby et al 2020), designed and built at CERN for the High Luminosity Project (Brüning and Rossi 2015). The CCT for High Luminosity LHC is the first application of CCT magnets in a collider. Superconducting CCT magnets as possible backbone of SC gantry for ions is the investigation topic of a large European collaboration, formed by CERN, CEA (Saclay, Fr), CIEMAT (Madrid), INFN (Milano and Genova), PSI (Ch), Uppsala University (Se), and Wigner Research Center for Physics (Hu). This collaboration has received two grants, one as part of the H2020-HITRIplus program (HITRIplus n.d.) and the other as part of the H2020-I.FAST program (I.FAST n.d.). Both programs started in spring 2021. HITRIplus aims at building a 4 T demonstrator of a CCT SC dipole for ion gantry, with the necessary curved shape, with Nb–Ti conductor. I.FAST comprises also three companies in the consortium, in addition to the above-mentioned laboratories: BNG (De), Elytt (Es), and Scanditronics (Se). I.FAST plans to test a straight CCT wound with HTS in the 4–5 T range. These programs are extensively reported (Rossi et al 2022a). It is worth noticing that in the frame of this collaboration a working group is considering how to adapt the field description used both by magnet designers and by beam optics designers, since the usual 2D field multipole decomposition is no longer valid for strongly curved magnets. In figure 5.45 a 3D view of the CCT magnet considered for HITRIplus is shown together with a cross-section of the coil-former package. In figure 5.45 it is shown that conductors lay in the grooves of the formers (the dashed sectors in the two annular formers) and that two solutions are being considered for the conductor shape: a Rutherford cable, similar to the one used to wind HEP magnets, and a rope cable, of the type 6 strands around a central core.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

HITRIPlus and IFAST should be able to assess in three years if CCT is a viable design to compete and maybe replace the classical cosϑ layout magnets.br

The cosϑ design—INFN SIG project

The above-mentioned collaboration among CERN, CNAO, INFN, and MedAustron is moving along two lines below described:

• (1)  
  CERN has refined the design of the 3 T dipole SIGRUM gantry and in (Karppinen et al 2022) a fairly detailed design is presented for a 70 mm aperture combined function 3 T dipole with superimposed a 5 T m⁻¹ gradient, capable of withstanding a ramp rate of 0.1 T s⁻¹. Figure 5.46, left side, shows a sketch of the magnet.
• (2)  
  INFN is investigating the possibility to improve the magnet performance, as recommended by the review panel. The study is supported by a 4-year grant of 1 M€ value that has been awarded by INFN management to study various technologies for the ion gantry, and had an effective start on the 1st of January 2022. The grant is complemented by contributions by CERN (250 k€) and CNAO (350 k€). The grant, called SIG, includes various items:

  • a.  
    Design, manufacture, and testing of a cosi dipole short prototype.
  • b.  
    Design of a scanning magnet system.
  • c.  
    Design and prototyping of the main components of a new dose-delivery system (DDS), integrated with a new range verification system (RVS) for ions.

The superconducting demonstrator is the most critical item of the SIG program, taking 600 k€ of the INFN budget, matched by the 600 k€ contribution of CNAO and CERN.

For SIG we are investigating the following parameters range (Rossi et al 2022b):

• 4–5 T dipole field, which entails a bending radius of 1.35–1.65 m.
• 70–90 mm coil diameter (aperture).
• 0.4 T s⁻¹ field continuous ramp rate up-down, with indirect cooling of the coils.

In figure 5.46, on the right side, a coil cross-section with field map for a possible SIG design is shown, together with a top view of the coil, well showing the 'banana shape'.

Zoom In Zoom Out Reset image size

The Tevatron, HERA, and LHC (at 4.2 K) dipoles feature similar field and aperture (only RHIC has a lower field) (Rossi and Bottura 2012, Bottura and Rossi 2015), but these colliders have a very low-field ramp (0.01–0.03 T s⁻¹). Only the INFN-Discorap prototype dipole (Fabbricatore et al 2011, Sorbi et al 2013), a built-in collaboration with GSI for FAIR-SIS300, has a faster ramp rate, at about 0.6–1 T s⁻¹, but requires direct cooling, with forced flow supercritical Helium.

The main challenge of the SIG prototype is the strong curvature, of 1.65 m. For comparison, DISCORAP has a 65 m bending radius, with the other projects featuring almost straight magnets.

The HIMAC gantry by Toshiba (Iwata et al 2012) is based on dipoles of 2.9 and 2.3 T, with a particular shape (curved race-track coils), with curvature radius of about 3 m. To go beyond this achievement of HIMAC the route of accelerator magnets, of tubular shape and rather low weight, seems the most promising one: with SIG 4 T dipoles, and a new mechanical structure under study at CERN, the weight of the gantry can be of 50–80 tons only.

To conclude this overview, we refer to the plot of figure 5.47, where magnets of different projects are reported in a chart where the squared product of field times the magnet aperture (this parameter gives the class of the magnet since it is proportional to the stored energy) is plotted versus the bending radius. Here the breakthrough made by the HIMAC magnets toward smaller bending radius, and the further step required by the SIG project both in bending radius and in stored energy, is rather evident in this magnet panorama.

Zoom In Zoom Out Reset image size

The SIG project, as part of the global SIGRUM collaboration, aims at testing the first prototype by 2025 and then to be able to proceed to the gantry construction. A possible sketch of the SIGRUM gantry installed in the CNAO facility is shown in figure 5.48.

Zoom In Zoom Out Reset image size

Superconducting magnets have accompanied the progress of HEP colliders. The community is pursuing the R&D on new more advanced technology for next-generation magnets for 15 T or more, necessary for the post-High Luminosity LHC collider. However, the same community is exploring if the presently available accelerator magnet technology can make a key contribution to cancer therapy, by enabling light and less expensive ion gantries. If successful, this development would give a significant boost to the performance of present hadron therapy centers based on heavy ions.