Table of contents

The aim is to develop and evaluate machine learning methods for generating quantitative parametric maps of effective atomic number (Z_eff), relative electron density (ρ_e), mean excitation energy (I_x), and relative stopping power (RSP) from clinical dual-energy CT data. The maps could be used for material identification and radiation dose calculation.

Machine learning methods of historical centroid (HC), random forest (RF), and artificial neural networks (ANN) were used to learn the relationship between dual-energy CT input data and ideal output parametric maps calculated for phantoms from the known compositions of 13 tissue substitutes. After training and model selection steps, the machine learning predictors were used to generate parametric maps from independent phantom and patient input data. Precision and accuracy were evaluated using the ideal maps. This process was repeated for a range of exposure doses, and performance was compared to that of the clinically-used dual-energy, physics-based method which served as the reference.

The machine learning methods generated more accurate and precise parametric maps than those obtained using the reference method. Their performance advantage was particularly evident when using data from the lowest exposure, one-fifth of a typical clinical abdomen CT acquisition. The RF method achieved the greatest accuracy. In comparison, the ANN method was only 1% less accurate but had much better computational efficiency than RF, being able to produce parametric maps in 15 s.

Machine learning methods outperformed the reference method in terms of accuracy and noise tolerance when generating parametric maps, encouraging further exploration of the techniques. Among the methods we evaluated, ANN is the most suitable for clinical use due to its combination of accuracy, excellent low-noise performance, and computational efficiency.

The recent interest in the integration of external beam radiotherapy with a magnetic resonance (MR) imaging unit offers the potential for real-time adaptive tumour tracking during radiation treatment. The tracking of large tumours which follow a rapid trajectory may best be served by the generation of a projection image from the perspective of the beam source, or 'beam's eye view' (BEV). This type of image projection represents the path of the radiation beam, thus enabling rapid compensations for target translations, rotations and deformations, as well time-dependent critical structure avoidance. MR units have been traditionally incapable of this type of imaging except through lengthy 3D acquisitions and ray tracing procedures. This work investigates some changes to the traditional MR scanner architecture that would permit the direct acquisition of a BEV image suitable for integration with external beam radiotherapy. Based on the theory presented in this work, a phantom was imaged with nonlinear encoding-gradient field patterns to demonstrate the technique. The phantom was constructed with agarose gel tubes spaced two cm apart at their base and oriented to converge towards an imaginary beam source 100 cm away. A corresponding virtual phantom was also created and subjected to the same encoding technique as in the physical demonstration, allowing the method to be tested without hardware limitations. The experimentally acquired and simulated images indicate the feasibility of the technique, showing a substantial amount of blur reduction in a diverging phantom compared to the conventional imaging geometry, particularly with the nonlinear gradients ideally implemented. The theory is developed to demonstrate that the method can be adapted in a number of different configurations to accommodate all proposed integration schemes for MR units and radiotherapy sources. Depending on the configuration, the implementation of this technique will require between two and four additional nonlinear encoding coils.

Differences in detector response between measured small fields, f_clin, and wider reference fields, f_msr, can be overcome by using correction factors or by designing detectors with field-size invariant responses. The changing response in small fields is caused by perturbations of the electron fluence within the detector sensitive volume. For solid-state detectors, it has recently been suggested that these perturbations might be caused by the non-water-equivalent effective atomic numbers Z of detector materials, rather than by their non-water-like densities. Using the EGSnrc Monte Carlo code we have analyzed the response of a PTW 60017 diode detector in a 6 MV beam, calculating the correction factor from computed doses absorbed by water and by the detector sensitive volume in 0.5  ×  0.5 and 4  ×  4 cm² fields. In addition to the 'real' detector, fully modelled according to the manufacturer's blue-prints, we calculated doses and factors for a 'Z  →  water' detector variant in which mass stopping-powers and microscopic interaction coefficients were set to those of water while preserving real material densities, and for a 'density  →  1' variant in which densities were set to 1 g cm⁻³, leaving mass stopping-powers and interaction coefficients at real levels. equalled 0.910  ±  0.005 (2 standard deviations) for the real detector, was insignificantly different at 0.912  ±  0.005 for the 'Z  →  H₂O' variant, but equalled 1.012  ±  0.006 for the 'density  →  1' variant. For the 60017 diode in a 6 MV beam, then, was determined primarily by the detector's density rather than its atomic composition. Further calculations showed this remained the case in a 15 MV beam. Interestingly, the sensitive volume electron fluence was perturbed more by detector atomic composition than by density; however, the density-dependent perturbation varied with field-size, whereas the Z-dependent perturbation was relatively constant, little affecting .

Treatment of small skin lesions using HDR brachytherapy applicators is a widely used technique. The shielded applicators currently available in clinical practice are based on a tungsten-alloy cup that collimates the source-emitted radiation into a small region, hence protecting nearby tissues. The goal of this manuscript is to evaluate the correction factors required for dose measurements with a plane-parallel ionization chamber typically used in clinical brachytherapy for the 'Valencia' and 'large field Valencia' shielded applicators. Monte Carlo simulations have been performed using the PENELOPE-2014 system to determine the absorbed dose deposited in a water phantom and in the chamber active volume with a Type A uncertainty of the order of 0.1%. The average energies of the photon spectra arriving at the surface of the water phantom differ by approximately 10%, being 384 keV for the 'Valencia' and 343 keV for the 'large field Valencia'. The ionization chamber correction factors have been obtained for both applicators using three methods, their values depending on the applicator being considered. Using a depth-independent global chamber perturbation correction factor and no shift of the effective point of measurement yields depth-dose differences of up to 1% for the 'Valencia' applicator. Calculations using a depth-dependent global perturbation factor, or a shift of the effective point of measurement combined with a constant partial perturbation factor, result in differences of about 0.1% for both applicators. The results emphasize the relevance of carrying out detailed Monte Carlo studies for each shielded brachytherapy applicator and ionization chamber.

In the abdomen, it is challenging to assess the accuracy of deformable image registration (DIR) for individual patients, due to the lack of clear anatomical landmarks, which can hamper clinical applications that require high accuracy DIR, such as adaptive radiotherapy. In this study, we propose and evaluate a methodology for estimating the impact of uncertainties in DIR on calculated accumulated dose in the upper abdomen, in order to aid decision making in adaptive treatment approaches.

Sixteen liver metastasis patients treated with SBRT were evaluated. Each patient had one planning and three daily treatment CT-scans. Each daily CT scan was deformably registered 132 times to the planning CT-scan, using a wide range of parameter settings for the registration algorithm. A subset of 'realistic' registrations was then objectively selected based on distances between mapped and target contours. The underlying 3D transformations of these registrations were used to assess the corresponding uncertainties in voxel positions, and delivered dose, with a focus on accumulated maximum doses in the hollow OARs, i.e. esophagus, stomach, and duodenum.

The number of realistic registrations varied from 5 to 109, depending on the patient, emphasizing the need for individualized registration parameters. Considering for all patients the realistic registrations, the 99th percentile of the voxel position uncertainties was 5.6  ±  3.3 mm. This translated into a variation (difference between 1st and 99th percentile) in accumulated D_max in hollow OARs of up to 3.3 Gy. For one patient a violation of the accumulated stomach dose outside the uncertainty band was detected.

The observed variation in accumulated doses in the OARs related to registration uncertainty, emphasizes the need to investigate the impact of this uncertainty for any DIR algorithm prior to clinical use for dose accumulation. The proposed method for assessing on an individual patient basis the impact of uncertainties in DIR on accumulated dose is in principle applicable for all DIR algorithms allowing variation in registration parameters.

The flexibility and sophistication of modern radiotherapy treatment planning and delivery methods have advanced techniques to improve the therapeutic ratio. Contemporary dose optimization and calculation algorithms facilitate radiotherapy plans which closely conform the three-dimensional dose distribution to the target, with beam shaping devices and image guided field targeting ensuring the fidelity and accuracy of treatment delivery. Ultimately, dose distribution conformity is limited by the maximum deliverable dose gradient; shallow dose gradients challenge techniques to deliver a tumoricidal radiation dose while minimizing dose to surrounding tissue. In this work, this 'dose delivery resolution' observation is rigorously formalized for a general dose delivery model based on the superposition of dose kernel primitives. It is proven that the spatial resolution of a delivered dose is bounded by the spatial frequency content of the underlying dose kernel, which in turn defines a lower bound in the minimization of a dose optimization objective function. In addition, it is shown that this optimization is penalized by a dose deposition strategy which enforces a constant relative phase (or constant spacing) between individual radiation beams. These results are further refined to provide a direct, analytic method to estimate the dose distribution arising from the minimization of such an optimization function. The efficacy of the overall framework is demonstrated on an image guided small animal microirradiator for a set of two-dimensional hypoxia guided dose prescriptions.

We evaluated the performance characteristics of a prototype preclinical PET scanner available as an easy clippable assembly that can dock to an MRI system. The single ring version of the PET system consists of eight detectors, each of which comprises a 12  ×  12 silicon photomultipliers (SiPMs) array coupled with a dual layer of offset scintillation crystals to measure depth of interaction. The crystal arrays have 29  ×  29 (30  ×  30 for the outer layer) 4 mm long LYSO crystals (6 mm for the outer layer). The ring diameter is 119.2 mm and the axial field of view is 50.4 mm. The NEMA NU 4-2008 protocol was followed for studying the PET performance. Temperature stability of SiPMs was also investigated. The peak system absolute sensitivity was 4.70% with an energy window of 250–750 keV. The spatial resolution was 1.28/1.88/1.85 mm FWHM (radial/tangential/axial) at a distance of 5 mm from the center. Peak noise equivalent counting rate and scatter fraction for mouse phantom were 61.9 kcps at 14.9 MBq and 21.0%, respectively. The uniformity was 6.3% and the spill-over ratios in the images of the water-and air-filled chambers were 0.07 and 0.17, respectively. Recovery coefficients ranged from 0.13 to 0.96. Change in sensitivity as a function of ambient temperature was 0.3%/°C. These first results indicate excellent spatial resolution performance for use with animal studies. Moreover, the clippable assembly can be upgraded to accept a second ring of SiPMs modules, leading to improved sensitivity and axial coverage.

A generic formalism is proposed for reference dosimetry in the presence of a magnetic field. Besides the regular correction factors from the conventional reference dosimetry formalisms, two factors are used to take into account magnetic field effects: (1) a dose conversion factor to correct for the change in local dose distribution and (2) a correction of the reading of the dosimeter used for the reference dosimetry measurements. The formalism was applied to the Elekta MRI-Linac, for which the 1.5 T magnetic field is orthogonal to the 7 MV photon beam. For this setup at reference conditions it was shown that the dose decreases with increasing magnetic field strength. The reduction in local dose for a 1.5 T transverse field, compared to no field is 0.51%  ±  0.03% at the reference point of 10 cm depth. The effect of the magnetic field on the reading of the dosimeter was measured for two waterproof ionization chambers types (PTW 30013 and IBA FC65-G) before and after multiple ramp-up and ramp-downs of the magnetic field. The chambers were aligned perpendicular and parallel to the magnetic field. The corrections of the readings of the perpendicularly aligned chambers were 0.967  ±  0.002 and 0.957  ±  0.002 for respectively the PTW and IBA ionization chambers. In the parallel alignment the corrections were small; 0.997  ±  0.001 and 1.002  ±  0.003 for the PTW and IBA chamber respectively. The change in reading due to the magnetic field can be measured by individual departments. The proposed formalism can be used to determine the correction factors needed to establish the absorbed dose in a magnetic field. It requires Monte Carlo simulations of the local dose and measurements of the response of the dosimeter. The formalism was successfully implemented for the MRI-Linac and is applicable for other field strengths and geometries.

Myocardial perfusion computed tomography (MPCT) imaging is commonly used to detect myocardial ischemia quantitatively. A limitation in MPCT is that an additional radiation dose is required compared to unenhanced CT due to its repeated dynamic data acquisition. Meanwhile, noise and streak artifacts in low-dose cases are the main factors that degrade the accuracy of quantifying myocardial ischemia and hamper the diagnostic utility of the filtered backprojection reconstructed MPCT images. Moreover, it is noted that the MPCT images are composed of a series of 2/3D images, which can be naturally regarded as a 3/4-order tensor, and the MPCT images are globally correlated along time and are sparse across space. To obtain higher fidelity ischemia from low-dose MPCT acquisitions quantitatively, we propose a robust statistical iterative MPCT image reconstruction algorithm by incorporating tensor total generalized variation (TTGV) regularization into a penalized weighted least-squares framework. Specifically, the TTGV regularization fuses the spatial correlation of the myocardial structure and the temporal continuation of the contrast agent intake during the perfusion. Then, an efficient iterative strategy is developed for the objective function optimization. Comprehensive evaluations have been conducted on a digital XCAT phantom and a preclinical porcine dataset regarding the accuracy of the reconstructed MPCT images, the quantitative differentiation of ischemia and the algorithm's robustness and efficiency.

Current clinical ultrasound scanners cannot be used to image the interior morphology of bones because these scanners fail to address the complicated physics involved for exact image reconstruction. Here, we show that if the physics is properly addressed, bone cortex can be imaged using a conventional transducer array and a programmable ultrasound scanner. We provide in vivo proof for this technique by scanning the radius and tibia of two healthy volunteers and comparing the thickness of the radius bone with high-resolution peripheral x-ray computed tomography. Our method assumes a medium that is composed of different homogeneous layers with unique elastic anisotropy and ultrasonic wave-speed values. The applicable values of these layers are found by optimizing image sharpness and intensity over a range of relevant values. In the algorithm of image reconstruction we take wave refraction between the layers into account using a ray-tracing technique. The estimated values of the ultrasonic wave-speed and anisotropy in cortical bone are in agreement with ex vivo studies reported in the literature. These parameters are of interest since they were proposed as biomarkers for cortical bone quality. In this paper we discuss the physics involved with ultrasound imaging of bone and provide an algorithm to successfully image the first segment of cortical bone.

Positron emission tomography (PET) is a functional imaging modality widely used in neuroscience studies. To obtain meaningful quantitative results from PET images, attenuation correction is necessary during image reconstruction. For PET/MR hybrid systems, PET attenuation is challenging as magnetic resonance (MR) images do not reflect attenuation coefficients directly. To address this issue, we present deep neural network methods to derive the continuous attenuation coefficients for brain PET imaging from MR images. With only Dixon MR images as the network input, the existing U-net structure was adopted and analysis using forty patient data sets shows it is superior to other Dixon-based methods. When both Dixon and zero echo time (ZTE) images are available, we have proposed a modified U-net structure, named GroupU-net, to efficiently make use of both Dixon and ZTE information through group convolution modules when the network goes deeper. Quantitative analysis based on fourteen real patient data sets demonstrates that both network approaches can perform better than the standard methods, and the proposed network structure can further reduce the PET quantification error compared to the U-net structure.

The purpose of this study is to examine in a clinical setting a novel formulation of objective functions for intensity-modulated radiotherapy treatment plan multicriteria optimization (MCO) that we suggested in a recent study. The proposed objective functions are extended with dynamic multileaf collimator (DMLC) delivery constraints from the literature, and a tailored interior point method is described to efficiently solve the resulting optimization formulation. In a numerical planning study involving three patient cases, DMLC plans Pareto optimal to the MCO formulation with the proposed objective functions are generated. Evaluated based on pre-defined plan quality indices, these DMLC plans are compared to conventionally generated DMLC plans. Comparable or superior plan quality is observed. Supported by these results, the proposed objective functions are argued to have a potential to streamline the planning process, since they are designed to overcome the methodological shortcomings associated with the conventional penalty-based objective functions assumed to cause the current need for time-consuming trial-and-error parameter tuning. In particular, the increased accuracy of the planning tools imposed by the proposed objective functions has the potential to make the planning process less complicated. These conclusions position the proposed formulation as an alternative to existing methods for automated planning.

Existing volumetric modulated arc therapy (VMAT) optimization using coplanar arcs is highly efficient but usually dosimetrically inferior to intensity modulated radiation therapy (IMRT) with optimized non-coplanar beams. To achieve both dosimetric quality and delivery efficiency, we proposed in this study, a novel integrated optimization method for non-coplanar VMAT (4πVMAT). 4πVMAT with direct aperture optimization (DAO) was achieved by utilizing a least square dose fidelity objective, along with an anisotropic total variation term for regularizing the fluence smoothness, a single segment term for imposing simple apertures, and a group sparsity term for selecting beam angles. Continuous gantry/couch angle trajectories were selected using the Dijkstra's algorithm, where the edge and node costs were determined based on the maximal gantry rotation speed and the estimated fluence map at the current iteration, respectively. The couch–gantry–patient collision space was calculated based on actual machine geometry and a human subject 3D surface. Beams leading to collision are excluded from the DAO and beam trajectory selection (BTS). An alternating optimization strategy was implemented to solve the integrated DAO and BTS problem. The feasibility of 4πVMAT using one full-arc or two full-arcs was tested on nine patients with brain, lung, or prostate cancer. The plan was compared against a coplanar VMAT (2πVMAT) plan using one additional arc and collimator rotation. Compared to 2πVMAT, 4πVMAT reduced the average maximum and mean organs-at-risk dose by 9.63% and 3.08% of the prescription dose with the same target coverage. R50 was reduced by 23.0%. Maximum doses to the dose limiting organs, such as the brainstem, the major vessels, and the proximal bronchus, were reduced by 8.1 Gy (64.8%), 16.3 Gy (41.5%), and 19.83 Gy (55.5%), respectively. The novel 4πVMAT approach affords efficient delivery of non-coplanar arc trajectories that lead to dosimetric improvements compared with coplanar VMAT using more arcs.

Vascular centerlines have crucial significance in reconstruction, registration, segmentation and vascular parameter analysis. The extraction of vessel structures remains a difficult problem in the completeness and continuity of results. In this paper, we present a novel method to extract cerebrovascular centerlines from four-dimensional computed tomography angiography images. Tubular features and vascular directions are used to extract initial centerlines, and the offset correction is introduced in the vascular orthogonal plane. In addition, we also present a post-processing method to connect interruptions of centerlines. We perform a quantitative validation using clinical images and public data sets of MRA brain images. Our experimental results demonstrate that the proposed algorithm not only shows higher accuracy in complicated vessel structures, but also outperforms previous approaches in terms of high validity and universality.

This is the second paper arising from a project concerning the application of Monte Carlo simulations to provide scanner-specific organ dose coefficients for modern CT scanners. The present focus is centred on the bone dosimetry models that have been developed. Simulations have been performed in photon only transport mode, with the assumption of electron equilibrium. This approximation breaks down for doses to active marrow and endosteum since the target cells are localised within tens of micrometre from bone tissue and dose enhancement functions are necessary to correct for the additional dose from photoelectric electrons created in adjacent material. The dose enhancement models used previously in publications NRPB-SR250 (Jones and Shrimpton 1993 Software Report NRPB-SR250, National Radiological Protection Board, Chilton, UK) and ORNL-TM8381 (Cristy and Eckerman 1987 Technical Report Oak Ridge National Laboratory, Oak Ridge, TN) have been implemented and compared with the contemporary approaches of Johnson et al (2011 Phys. Med. Biol. 56 2347–65) and ICRP Publication 116 (ICRP 2010 Ann. ICRP40 1–257) that are being adopted in the present project. In addition, the calculation of dose to endosteum in the medullary cavity is reviewed and updated using electron mode simulations. For the purposes of quality assurance and comparison, the various dose enhancement functions have been applied in relation to the NRPB18+DJ and HPA18+  stylised hermaphrodite phantoms and also the adult male and female voxel phantoms recommended in ICRP Publication 110 (ICRP 2009 Ann. ICRP39 1–165), for exposure from three CT scanners modelled previously. Contemporary results for standard examinations on the head and trunk calculated for these latter phantoms demonstrate moderate increases (modal value  +18%) in active marrow dose coefficients relative to values derived from data published in NRPB-SR250. A similar analysis in relation to endosteum dose coefficients shows larger reductions (modal value  −46%), owing at least in part to changes in assumed location of the target cells. Even larger changes are apparent for both of these dose coefficients in relation to examination of the upper legs (−39% and  −94%, respectively). However, resultant changes in any values of effective dose will be less owing to the low weighting factors applied for these tissues.

Megavoltage (MV) cone-beam computed tomography (CBCT) using an electronic portal imaging (EPID) offers advantageous features, including 3D mapping, treatment beam registration, high-z artifact suppression, and direct radiation dose calculation. Adoption has been slowed by image quality limitations and concerns about imaging dose. Developments in imager design, including pixelated scintillators, structured phosphors, inexpensive scintillation materials, and multi-layer imager (MLI) architecture have been explored to improve EPID image quality and reduce imaging dose. The present study employs a hybrid Monte Carlo and linear systems model to determine the effect of detector design elements, such as multi-layer architecture and scintillation materials. We follow metrics of image quality including modulation transfer function (MTF) and noise power spectrum (NPS) from projection images to 3D reconstructions to in-plane slices and apply a task based figure-of-merit, the ideal observer signal-to-noise ratio (d') to determine the effect of detector design on object detectability. Generally, detectability was limited by detector noise performance. Deploying an MLI imager with a single scintillation material for all layers yields improvement in noise performance and d' linear with the number of layers. In general, improving x-ray absorption using thicker scintillators results in improved DQE(0). However, if light yield is low, performance will be affected by electronic noise at relatively high doses, resulting in rapid image quality degradation. Maximizing image quality in a heterogenous MLI detector (i.e. multiple different scintillation materials) is most affected by limiting total noise. However, while a second-order effect, maximizing total spatial resolution of the MLI detector is a balance between the intensity contribution of each layer against its individual MTF. So, while a thinner scintillator may yield a maximal individual-layer MTF, its quantum efficiency will be relatively low in comparison to a thicker scintillator and thus, intensity contribution may be insufficient to noticeably improve the total detector MTF.

Photo-mediated ultrasound therapy (PUT) is a non-invasive, agent-free technique to shut down microvessels with high precision by promoting cavitation activity precisely in the targeted microvessels. PUT is based on the photoacoustic (PA) cavitation generated through concurrently applied nanosecond laser pulses and ultrasound bursts. In this study, a PA cavitation model is employed to understand the enhanced cavitation activity during PUT, with full consideration of the optical absorption of blood vessels. Bubble size evolution in cylindrically-shaped optical absorbers (vessels) due to rectified diffusion is simulated. Results show that the ultrasound pressure required for bubble growth decreases dramatically with the increased laser fluence. At a relatively low ultrasound driving pressure, bubble equilibrium radius increases rapidly due to concurrently applied nanosecond laser pulses and ultrasound bursts, resulting in a transition from inertial cavitation to stable cavitation. This inertial to stable transition is verified by the experimentally measured results on 0.76 mm silicone tubes filled with human whole blood with 0.5 MHz ultrasound at 0.243 MPa. This study demonstrated the potential to induce stable bubbles in blood vessels by PUT non-invasively.

The out-of-field dose in radiation therapy is a growing concern in regards to the late side-effects and secondary cancer induction. In high-energy x-ray therapy, the secondary neutrons generated through photonuclear reactions in the accelerator are part of this secondary dose. The neutron dose is currently not estimated by the treatment planning system while it appears to be preponderant for distances greater than 50 cm from the isocenter. Monte Carlo simulation has become the gold standard for accurately calculating the neutron dose under specific treatment conditions but the method is also known for having a slow statistical convergence, which makes it difficult to be used on a clinical basis. The neutron track length estimator, a neutron variance reduction technique inspired by the track length estimator method has thus been developped for the first time in the Monte Carlo code GATE to allow a fast computation of the neutron dose in radiotherapy. The details of its implementation, as well as the comparison of its performances against the analog MC method, are presented here. A gain of time from 15 to 400 can be obtained by our method, with a mean difference in the dose calculation of about 1% in comparison with the analog MC method.

Although luminescence of water lower in energy than the Cerenkov-light threshold during proton and carbon-ion irradiation has been found, the phenomenon has not yet been implemented for Monte Carlo simulations. The results provided by the simulations lead to misunderstandings of the physical phenomenon in optical imaging of water during proton and carbon-ion irradiation. To solve the problems, as well as to clarify the light production of the luminescence of water, we modified a Monte Carlo simulation code to include the light production from the luminescence of water and compared them with the experimental results of luminescence imaging of water. We used GEANT4 for the simulation of emitted light from water during proton and carbon-ion irradiation. We used the light production from the luminescence of water using the scintillation process in GEANT4 while those of Cerenkov light from the secondary electrons and prompt gamma photons in water were also included in the simulation. The modified simulation results showed similar depth profiles to those of the measured data for both proton and carbon-ion. When the light production of 0.1 photons/MeV was used for the luminescence of water in the simulation, the simulated depth profiles showed the best match to those of the measured results for both the proton and carbon-ion compared with those used for smaller and larger numbers of photons/MeV. We could successively obtain the simulated depth profiles that were basically the same as the experimental data by using GEANT4 when we assumed the light production by the luminescence of water. Our results confirmed that the inclusion of the luminescence of water in Monte Carlo simulation is indispensable to calculate the precise light distribution in water during irradiation of proton and carbon-ion.

The MR-Linac will provide excellent soft tissue contrast for on-treatment imaging. It is well known that the electron return effect (ERE) results in areas of increased and decreased dose at air/tissue boundaries, which can be compensated for in plan optimisation. However, anatomical changes may affect the quality of this compensation. In this paper we aim to quantify the interaction of anatomical changes with ERE in head and neck (H&N) cancer patients.

Twenty patients treated with either 66 Gy or 60 Gy in 30 fractions were selected. Ten had significant weight-loss during treatment requiring repeat CT (rCT) and ten had PTVs close to the sinus cavity. Plans were optimised using Monaco to meet the departmental dose constraints and copied to the rCT and re-calculated. For the sinus patients, we optimised plans with full and empty sinus at both 0 T and 1.5 T. The effect of the opposite filling state was next evaluated.

No clinically relevant difference between the doses in the PTV and OARs were observed related to weight-loss in 0 T or 1.5 T fields. Variable sinus filling caused greater dosimetric differences near the walls of the sinus for plans optimised with a full cavity in 1.5 T, indicating that optimising with an empty sinus makes the plan more robust to changes in filling. These findings indicate that current off-line strategies for adaptive planning for H&N patients are also valid on an MR-linac, if care is taken with sinus filling.

Multi atlas based segmentation (MABS) uses a database of atlas images, and an atlas selection process is used to choose an atlas subset for registration and voting. In the current state of the art, atlases are chosen according to a similarity criterion between the target subject and each atlas in the database. In this paper, we propose a new concept for atlas selection that relies on selecting the best performing group of atlases rather than the group of highest scoring individual atlases.

Experiments were performed using CT images of 50 patients, with contours of brainstem and parotid glands. The dataset was randomly split into two groups: 20 volumes were used as an atlas database and 30 served as target subjects for testing. Classic oracle selection, where atlases are chosen by the highest dice similarity coefficient (DSC) with the target, was performed. This was compared to oracle group selection, where all the combinations of atlas subgroups were considered and scored by computing DSC with the target subject. Subsequently, convolutional neural networks were designed to predict the best group of atlases. The results were also compared with the selection strategy based on normalized mutual information (NMI).

Oracle group was proven to be significantly better than classic oracle selection (p  <  10⁻⁵). Atlas group selection led to a median  ±  interquartile DSC of 0.740  ±  0.084, 0.718  ±  0.086 and 0.670  ±  0.097 for brainstem and left/right parotid glands respectively, outperforming NMI selection 0.676  ±  0.113, 0.632  ±  0.104 and 0.606  ±  0.118 (p  <  0.001) as well as classic oracle selection.

The implemented methodology is a proof of principle that selecting the atlases by considering the performance of the entire group of atlases instead of each single atlas leads to higher segmentation accuracy, being even better then current oracle strategy. This finding opens a new discussion about the most appropriate atlas selection criterion for MABS.