Review—Micro-Fuel Cell Principal Biosensors for Monitoring Transdermal Volatile Organic Compounds in Humans

Knowledge of transduction mechanisms in biosensing applications paves the way for ultrasensitive and dynamic detection in living systems. Real-world biosensing applications where ultra-sensitivity and dynamic detection are paramount include monitoring the anesthetic agent concentration during surgery; the slightest variation in concentration can potentially result in a life-threatening overdose or, on the other end of the spectrum, the patient’s awareness during the procedure. We review the benefits and functions of the transcutaneous biosensor device compared with other current technology and discuss the sensor’s capability to accurately measure volatile anesthetic gas concentration in blood using fuel cell technology. We review fundamental concepts of fuel-cell technology for wearable bio-sensing applications. The fuel cell sensor can also continuously monitor other volatile organic compounds making it versatile with numerous potential applications.

Scientists, researchers, and engineers have attempted to address the need for portable chemical sensors dating back to the 17th century. Historian J. Smart chronologically details a list of top highlights in the progression of nuclear/biological/chemical (NBC) threat agent analysis and the development of U.S. Army detectors, alarms, and warning systems dating back to World War I. Chemical sensing has progressed from chemical vapor detectors to chemical spectrum analysis and into the promising age of electrochemical sensing. The underlying concept behind electrochemical sensing technology is the transduction (and often amplification) of a chemical reaction into useable amperometric, potentiostatic, or conductometric electrical signals. The mechanisms of these classic electrical sensors, modern devices, and documentation of numerous applications have been thoroughly detailed throughout many review articles. [1][2][3] Electrical output from electrochemical sensors opens opportunities in wireless data transmission and continuous data monitoring. Furthermore, electrical outputs allow for machine learning analytics as displayed in Fig. 1 where classic sensing is compared to a machine learningenabled sensor. Machine Learning (ML) is a tool to be leveraged for predictive maintenance, non-linear calibration of electrochemical sensors, better sensitivity, and higher selectivity, especially with sensor arrays. Most notably ML-enabled sensors can relearn and adjust the model allowing increased accuracy over time as visualized in Fig. 1c of the machine learning cycle which is referenced against unseen data for validation. 4 Through the historical development of chemical sensors, a fundamental challenge is the collection of the analyte. Proper sample collection and analyte processing are critical to the accuracy of the diagnostic. Historically, chemical analyte collection and preparation has been a three-step process: sample collection, analyte isolation, and pre-concentration. It is a highly controlled process, time-consuming, and often requires additional solutions and buffers to process the analysis. The need for excess chemicals in the sensing process is not supportive of the "green chemistry" movement that is paramount for conserving the earth and preserving resources. Additionally, these methods are not realistic for portable, wearable technologies, and are not userfriendly for non-clinical applications. 5 One concept that combines sample collection, analyte isolation from sampled media, and pre-concentration in a single step without excess chemicals is passive diffusion. The sample media could be air, water, or across the skin (air and sweat) and even soil. This method is described in detail by Górecki and Namienik but given the importance of this concept, we highlight some key concepts. Passive diffusion is the sampling of freely flowing analyte from the sample media to a collection media by way of diffusion through a membrane (or another type of barrier). This works in a unidirectional flow so that the analyte is effectively trapped by the collecting media (Fig. 2). 5 A note on the limitation of this sampling method is the net flow will continue until it reaches an equilibrium (or until the sampling session is ended by the user). This equilibrium is the effective concentration upper limit. Awareness of the upper limit is important for real-world applications. Figure 3 shows an example of this. If the need for a sensor was to read ammonia within a certain expected range of concentrations (e.g., 0.5-1.0 mM ammonia or 8-17 ppm ammonia) the limitation of the sensor (ruled by this equilibria point) must be higher than the upper limit of the needed reading. The sensor data on the upper right side demonstrates a max reading of 1.5 mM concentration of ammonia, so this sensor would be suitable for the proposed scenario. On the left side, is the diagram sensor set-up design. It uses a membrane on the bottom with an internal electrolyte solution. We can note that in general, this configuration would be problematic for wearable applications as orientation and movement would greatly affect the stability and reliability of the sensor. 6 An approach to tackling equilibrium limitations is to use a non-passive diffusion method: pumping the air through the device. However, for wearable sensors, a pump would be too much of a penalty for a lowprofile sleek design as well as too much power consumption hurting the battery life. That said, we will not discuss this approach further since it is out of the scope of wearable design and we will continue to focus on passive diffusion-based sensors.
Mathematically, Fick's first law of diffusion can be used to describe the diffusion of an amount of analyte M, across the membrane in a certain amount of time, t(s), (when the concentration gradient is linear and the collection efficiency is 100%) according to the following equation: where U is the diffusive transport rate (mol/s), D is the molecular diffusion coefficient of the analyte (cm 2 /s), A is the cross-section of the diffusion path (cm 2 ), L is the total length of the diffusion path (cm) and c 0 is analyte concentration in the medium examined (mol/cm 3 ). 5 Fick's first law of diffusion is also applicable for describing gas diffusion across a membrane as follows: Where M, A, and t have the same meaning as above, S is the permeability of coefficient of a given analyte (cm 2 /min), L M is membrane thickness (cm), and p 1 is the partial pressure of the analyte near the external membrane surface. 5 A review by Li et. al discusses diffusion across a membrane for gas-sensing applications. As pictured in Fig. 4 they introduce different possible diffusion kinetic mechanisms: bulk Poiseuille flow, Knudsen diffusion, size-restricted diffusion, and solid-state diffusion. For sensor design, considerations must be made for the analyte molecule being detected. If it is small enough the sizerestricted or solid-state diffusion mechanisms could be beneficial for the selectivity of the sensor. However, if the molecule is larger it is important to be aware that other molecules could also diffuse across   the membrane in a bulk Poiseuille flow or Knudsen-type diffusion which may pose a challenge in selectivity. 7 Three configurations of the membrane-based gas chemical sensors reviewed by Li are visualized in Fig. 5. The membrane's functional role in these sensors is as a barrier to create a diffusion gradient and possibly to filter out larger molecules, but do not contribute to the sensing recognition, nor the signal transduction. Differently, in a fuel cell, the membrane is a functional component in the recognition and transduction processes. 7 Proton Exchange Membrane Fuel Cells (PEMFC) are energy conversion devices that efficiently transform chemical energy into electrical energy (see Fig. 6). The ion-selective membrane allows the anion to pass through, forcing the charge-caring electron to the outer circuit. Rather than employing the PEMFC capability for energy generation, this feature is leveraged as a sensing signal output, setting a foundation for a new class of chemical and electronic sensors: fuel cell-based. In fuel-cell-based sensors, the membrane contributes greatly to the simultaneous molecule identification and transduction of the sensing signal from the chemical realm into the desired electrical signal. The most challenging aspect of fuel cellbased sensors is the selectivity against interferents since the chemical recognition component is not separated from the membrane as it is for the classic electrochemical sensors. Additionally, the device configuration of fuel cell-based sensors is robust for wearable applications against expected disturbances from much movement and different positioned orientations.
Micro-fuel cell-based sensors are up-and-coming point-of-care devices, but there are still several challenges that impede the accuracy of portable point-of-care sensors. The first challenge, regardless of the sensing mechanism type, is an uncontrolled sensing environment. Sensors must be resilient against fluctuations in external temperatures, humidity, altitude, and other noises (e.g., mechanical vibrations in a medevac helicopter or movement of a patient's wrist in watch-style sensors). Humidity interference brings many concerns and is recognized as a major interferent for all transdermal sensing, especially so for FCs. Portable sensors are needed in high-consequence situations, including life or death scenarios such as warfare, medical, and forensic applications. Therefore, regardless of the challenge of an unpredictable environment, the device must be reliable and accurate as it will most likely be the only available device to perform on-site analysis.
The second challenge is selectivity and sensitivity toward the target itself (ion or chemical compound). These are the two classic sensor challenges. Selectivity is to sense only the target without reacting to interferents. Sensitivity is how much the signal will change proportional to the change in concentration of the target (the steeper the slope the better).
Classical approaches to these challenges include material selection, device configurations, and sensing amplification. Dr. Bhansali's interdisciplinary team has studied this innovative fuel cell sensor throughout the past few years, 9 using it to measure transdermal ethanol content in blood samples. The cells' membrane electrode assembly (MEA) consists of two individual catalyst electrodes on two sides separated by a proton exchange polymer electrolyte. To improve the redox reaction of the sandwiched configuration, the team investigated the effects of using different catalyst materials. The explorations of a wide variety of catalyst materials are reported throughout the literature, including metals (e.g., Pt, Ru, Pd, Sn, Au,  Mo, Cd, Bi, Ti), metal alloys, 10 nanocomposite materials, [11][12][13] and polymer matrices. 14 In detail, different VOCs require suitable catalysts for the most efficient catalysis behavior. For instance, regarding ethanol oxidation, Pt is considered the most efficient catalyst as the anode side of the MEA. 15 Such Material Engineering Research and Nanofabrication developments in fuel cell-based sensors substantially improve selectivity towards the target compound in the presence of interferents. Interferent elements pose challenges for all sensors. Commercially available sensors (e.g., breathalyzers, colorimetric sweat swabs) may occasionally result in "false positive" errors due to interferents. [16][17][18] Thus the selectivity, simplicity, stability, sensitivity, longevity, low cost, and low power consumption attributes of Fuel-Cell sensors demonstrate potential for surpassing existing volatile organic compound (VOC) monitoring devices in adoption into global healthcare.
Supplementing classical approaches, a cutting-edge approach for further optimizing selectivity and sensitivity is with machine learning. ML-enabled sensors. It may allow for improved selectivity by categorical analysis predictions. It could also improve sensitivity with regression analysis predictions. Simply put, categorical machine learning aims to properly identify a target or non-target (hence category), while regression-based machine learning aims to predict a numerical output more accurately (e.g., higher accuracy in the concentration detected). Beyond Machine learning techniques are deep-learning (DL) methods such as multimodal or multivariate analysis. DL helps find and identify unseen patterns in a vast amount of data that is often overlooked by user analysis. These techniques can help address concerns of drift, signal overlap, poor calibration, and false readings. Linking larger data sets gives robust fitting and establishes connections between the input variable: the concentration of VOC and the output variable: the electrical current. Another contribution of DL is to overcome the difficulties of calibration for finding concentration levels below ppm of analytes by electrochemical sensor research.
Another challenge for point-of-care sensors is patient comfort. Many invasive (implantable or skin penetrating) sensors are critical to saving lives, such as the invasive glucose monitoring sensors that penetrate the patient's skin. The blood glucose test consists of a twice-a-day finger pricking to monitor glucose. Continuous glucose monitoring (CGM) consists of a device embedded into the skin. CGM reduces the pricking of a needle from twice a day down to once per 10 days. 19 It also shows better time in a range which is a key parameter for ensuring steady glucose levels (neither too high nor too low) in the patient. Occasionally these CGM sensors can falter and if an inaccuracy is suspected it is then recommended to rely on a classic blood prick to confirm. These sensing methods are painful and unpleasant for patients. To clarify, a CGM has a separate skin insertion location from the insulin delivery pump insertion. Therefore, the patient is enduring two invasive ports on their body. This may be unavoidable for the insulin delivery system. However, there is potential to transform the CGM into a non-invasive sensing format (as will be discussed in detail later).
One category of non-invasive sensors is breathalyzers. Breathalyzers are commonly used by law enforcement for diagnosing blood alcohol content (BAC) in intoxicated drivers. 20 BAC breathalyzers are even available for purchase in the public marketplace for personal use for monitoring alcohol. 21 The challenges and drawbacks of current BAC breathalyzers are that (1) they are not continuously monitoring, (2) they are dependent on many factors including the rate and quantity of exhaled breath from the user, (3) suffer from the presence of many interferents (i.e., other particles in the breath.) The breathalyzers used by law enforcement are known to be more expensive but more reliable than the publicly available BAC sensors. 22 This poses a more dangerous challenge in that people are relying on their non-clinical BAC sensors, and these sensors may result in false readings below the legal limit which signifies it is safe to operate a vehicle when they are really over the legal limit. This could result in dire outcomes such as arrest, car accidents, or even death.
There is a great need for continuous monitoring sensors. Continuous monitoring provides better more instantaneous monitoring where high control is needed. Breathalyzers are also used for non-invasive general anesthetic monitoring and continuous monitoring of the patient. For instance, existing infrared analyzers (I.R.) for anesthesia must be operated with high-precision optical alignments which are costly, bulky, and complex. 23 These reasons cause low deployment of I.R. for critical healthcare, especially in communities with scarce resources. Uncontrolled anesthetic administration can bring severe adverse health outcomes to patients. Presenting a critical need for portable, low-cost, continuously monitoring, reliable VOC sensors.
Beyond the clinical settings, continuous monitoring has become popular and insightful for long-term patient health and early diagnosis that can be lifesaving. To date, some of the most advanced and popular wearable health technology that is also widely adopted by the public are the Apple, Fitbit, and Samsung smartwatches. Looking at the Apple watches, these sleek advanced devices are equipped with numerous sensors including Heartrate, Oximetry (oxygen in the blood), Temperature (Ambient and Skin), Barometric (Altitude), and Galvanometric (Skin Conductance) sensors. 24,25 Not to mention, a high level of computing capability with long (all-day) battery life. The next natural progression for electrochemical-based wearables is to follow the success of current watch-style heart monitoring devices with advanced sensing capability for VOC detection. As such, a continuously-monitoring, accurate, low-cost Blood Alcohol Content (BAC) sensor with successful market penetration would have a significant positive social impact on preventing the overconsumption of alcohol and the subsequent consequences.
In recent years PEMFC sensors have been employed to quantify an assortment of VOCs in humans for various diagnostics. Specific VOCs can be drawn as biomarkers for various diseases, cancers, and more. Delayed awareness of the disease or overexposure to hazardous VOCs could prove fatal. Therefore, real-time, continuous, non-invasive, cost-effective, and user-friendly sensing technology will significantly improve human healthcare conditions. Wearable fuel cell anesthesia biosensors can accurately measure the VOC of isoflurane gas concentration <40 ppm at a much lower cost/unit aimed to support global access to quality healthcare. 26 Battery life is another top challenge for point-of-care devices. If you need to charge your wearable sensor every 3 h it becomes useless for continuous monitoring and a nuisance for the user. Additionally, using a bigger batter (larger capacity) is not an option to maintain the sleek low-profile look that real-world users will not sacrifice. Power awareness for portable sensor design can even affect the three sensor traits that determine sensor viability: limit of detection, sensitivity, and selectivity. Power design in the circuit must be optimized to still allow for accurate data collection. If the sampling rate is not high enough (to preserve power consumption) then valuable data could be missed. Application-specific analysis should be conducted to understand the tradeoffs and optimize the device accordingly. 27 To reiterate, both circuit design and firmware should be designed to maintain the highest accuracy with the least battery consumption possible. "Power-aware design" as it's called, is critical to implement from the earliest stage of the design process and must not be overlooked.
To further support progress in the wearable domain a major challenge is the miniaturization of all components including the circuitry. A printed circuit board collects, processes, and transmits the data collected. The processing capabilities of the circuitry must be comparable to the processing capabilities of table-top in-lab equipment for accuracy and reliability in the results. Fuel cell-based sensors, like other classic electrochemical sensors, are analyzed by all the classic electrochemical characterization techniques including but not limited to cyclic voltammetry, chronoamperometry, electrochemical impedance spectroscopy, and open circuit potential. The bench-top potentiostat can perform these characterization techniques in-lab. In support of the scientific progression of wearables, Hoilett et al., researchers from the Biomedical and Electrical & Computer Engineering departments of Purdue University have reported that KickStat's response was within 6% of the Bio-logic VSP-300 Benchtop potentiostat. Furthermore, they have made both the design specs regarding their miniaturized amperometric circuit KickStat and the firmware publicly available on GitHub. 28 Their circuit design relies on the component of the LMP9100 potentiostat chip by Texas Instruments. Preceding their work by six years, a team from Florida International University had also employed the LMP91000 for sensing device design. Cruz et al. published their findings about the capabilities of the low-cost mini-potentiostat for point-of-care devices. 29 More than just miniaturization, to obtain a truly low-profile, skin-conforming wearable sensor the packaging for circuits and displays needs to be flexible. This would allow wearable sensors to flood the private market in the form of "bracelet-style" sensors over the current trending "watch-style" sensor configuration. Not only is this would be aesthetically pleasing to the public, but it could also support the pediatric realm of wearable sensors. Like the teams from Purdue and Florida Internal University, true Interdisciplinary Collaborative Research is the next big push in wearable point-of-care sensor advancement. Chemical Engineering, Material Engineering, Nanofabrication, Electrical Engineering, Bio-Engineering, and other areas of expertise will all support the progression of portable, highly sensitive, and selective devices with the capability to analyze the results on-site and even share results wirelessly.
In the aim for (1) aesthetically pleasing appearance, (2) low-cost, (3) low-power, (4) non-invasive, (5) wearable, (6) continuous monitoring, (7) and an easy to use (device and software) VOC sensor solution many of the discussed technologies have not been able to check all the boxes (Table I). These features cannot sacrifice the selectivity, sensitivity, and stability that are critical for sensors. Thus, we highlight the benefit of fuel cell (FC) wearables for pointof-care sensing technology.
The main emphasis of this work is (1) to introduce micro-fuel cell sensor-based transcutaneous anesthesia monitoring systems (TAMS), (2) to review and summarize recent research and progress comparing TAMS with other types of wearable VOC sensors, (3) to outline and emphasize the need for modern electrical perspectives including Machine learning for advanced sensing capabilities and power awareness for successful deployment of portable sensors.

Theory of PEMFC for VOC Sensing
Most current VOC sensors (Fig. 7) suffer from instability, nonlinearity, poor selectivity, and limited sensitivity at a lower concentration (a high cost in real-world applications). Electrochemical sensors are the most suitable portable devices for continuous real-time monitoring; Fuel cell sensors offer simplicity, high accuracy, and portability. 32 Electrochemistry for sensing.-Electrochemistry is an effective method for VOC detection. The technique is based on the principle of redox reactions. When the target analyte undergoes these redox reactions an output signal is produced as a direct correlation to the concentration of VOC. Previously shown in the Fig. 5 was a demonstration of amperometric, potentiometric, and conductometric configurations of electrochemical sensors. The advantage of electrochemical sensing is fast response time, high accuracy, a wider range of detection, and low power consumption.
Potentiometric sensors function by measuring the variation of potential difference (voltage) between working and reference electrodes under the conditions of no current flow. The sensor is validated in various concentrations to determine the linear response of the sensor (voltage as a function of concentration). This linearity of the sensor shows the sensitivity level of the device. 33 The signal of the potential is based on the Nernst equation. 34 Wherein the symbols are as follows. E: half-cell potential, E 0 : the standard potential, R: the molar gas constant, F: Faraday constant, n: the number of electrons, T: temperature, Q r : reaction quotient. Based on the Nernst equation, the current is dependent on the applied voltage, the open-circuit potential (OCP), and the reaction speed. The potential can then be used to determine the analytical quantity of interest e.g., the concentration of the target analyte.
In amperometric sensors, an optimal potential is applied and a corresponding current is obtained due to a reduction or oxidation reaction. Amperometric are like potentiometric sensors in that they are characterized by their current-potential relationship with the electrochemical system. The relationship between the current and the concentration of the analyte can be expressed by the Cottrell equation: Where I is current, z is the number of transferred electrons, F is the Faraday constant, A the surface area of the electrode, c* is the bulk concentration, D is the diffusion coefficient, and t is the time. 35 An impedance sensor works by measuring the change in impedance between the two electrodes of a sensor while applying a sinusoidal voltage; the change in impedance is calculated from the measured current. 36 The sensing response is evaluated from subhertz to mega-hertz. The electrochemical impedance spectroscopy allows in-depth insight into the sensing system for high sensitivity and delineates the complexity of the change in the sensing environment, such as a change in concentration, leading to a corresponding variation in the sensor's impedance. Impedance is comprised of contributions from inductance, capacitance, and resistance. The combination of all can be explained by the complex impedance itself. Electrochemical impedance spectroscopy (EIS) is ideally suited for the analysis of small amplitude stimulus to an interface (high sensitivity to change in concentrations) by leveraging the baseline measurements performed in a steady-state condition. The spectral analysis allows the isolation of different time constants specific to each contribution to the global mechanism, including double-layer charging, electrochemical charge transfer, mass transport, and adsorption. Figure 8 shows that EIS can be plotted in the Warburg plot to analyze the diffusion mechanisms in an electrochemical system such as an electrochemical-based sensor. 37 Accessibility and stability of biological elements.-In his review "Volatile Metabolites" Rowan explains the nature of VOCs to have   concentration levels in a patient are examples of the applications of exogenous VOC detection. Glucose monitoring and cancer screening are applications for endogenous VOC monitoring. VOCs can be detected in breath and biofluids (blood, urine, feces, sweat, and interstitial fluid). Previously documented by Anand et al. were some methods for specimen collection and the stability of trace element analysis. 39 They describe how the integrity of the specimen could be compromised even before it is analyzed, by contamination during collection and processing, or by attenuation of the analyte concentration during storage. They highlight the importance of controlled procedures at every stage to mitigate such errors. Continuous monitoring with wearable sensors addresses many challenges in the stability of biofluids for sensing. Sensing the analyte at the source and acquiring the data uninterrupted manner allows for more reliability in the data.
The stability of biological elements in the various sensing mediums is a great concern in biosensing. Blood samples have high reliability, but this technique is invasive and not ideal for patients. The reliability of Breathalyzer-based sensors is quite dependent on the quantity and momentum of exhaled breath. So, although breathalyzers are non-invasive, reliability remains a challenge. Breathalyzers can also suffer from biofouling (contamination from food and other interferents). Furthermore, the VOCs detected in breath vary greatly from patient to patient making a sensing base point more challenging. One case study on the VOCs emitted by breath collected samples from 50 subjects. In total 3,481 unique VOCs were identified, yet among those only 27 VOCs were observed in all 50 subjects. 40 For the non-invasive transcutaneous FC sensors, the stability of the VOCs is already much more reliable than breathalyzers. Transcutaneous sensors are either sweat based or dependent on the interstitial fluid (fluid between cells also called tissue fluid). Sweat is either induced through activity or medicinal ways. The availability of sweat also varies between patients and for some neither exercise nor medicinal methods may be an option (considering the elderly, ill, disabled, and those with allergies, etc). 31 Interstitial fluid can either be extracted by microneedles or with an external voltage applied across the skin. Microneedles are still invasive (albeit minorly) and are not ideal for continuous monitoring applications. The external voltage is completely non-invasive working via the concepts of ions and an electrical field. When the electric field is applied the positively charged ions in the interstitial fluid are attracted into the field and drawn upward and out of the skin. Not only are these ions drawn out, but they have been found to carry out other noncharged species found in the interstitial fluid. 31 Demonstrated in Fig. 9 is an example of a classic three-electrodebased electrochemical sensor enhanced with the added extraction electrodes for a non-invasive interstitial fluid-based sensor design. They present the watch-style configuration, circuit board, electrode diagram, cross-sectional diagram of interstitial fluid extraction from the skin, and the electrical flow chart to phone-displayed data. The double-working electrode configuration acts as the smallest sensor network to further enhance sensitivity and stability. Furthermore, this design supports the continuous monitoring landscape of sensors. By simply modulating between extraction and sensing protocols in the software the sensor can continuously monitor glucose in the patient. 31 In the past Fuel cell-based sensors have been used to monitor interstitial fluid as implantable sensors. 41 A novel approach to support non-invasive continuous monitoring of biofluids would be to incorporate extraction electrodes into a fuel cell-based sensing system to extract interstitial fluid, inspired by the work of Chang et al. 31 Anesthesia.-As discussed, it is critical to monitor anesthesia levels for patients under operation. Low-resource areas need equipment with reliable monitoring at an accessible price point. To address this, Fuel Cell design can be employed in Transcutaneous Anesthesia Monitoring Systems (TAMS). The widely used anesthesia, isoflurane, undergoes a minimal metabolism after being inhaled. The excess VOCs are exhaled and excreted by sensible/ insensible perspiration. 26 Considering the minimal metabolism of excess VOCs, the concentration levels detected are positively proportional to the patient-absorbed concentrations of isoflurane. Quantifying transdermal isoflurane vapor emissions and defining the exact correlation between the measured VOC with the patientabsorbed isoflurane enables direct monitoring of anesthetic levels. The reaction mechanism includes VOC oxidation reactions and oxygen reduction reactions. The additive oxidation reactions at the anode are written in Eqs. 5-7. The oxidative addition of isoflurane takes place rather than the single-step reaction. 26 Wherein the R-Cl refers to C 3 H 2 F 5 O-Cl. At the cathode, the oxygen reduction is as below (see Fig. 10).
A miniaturized portable PEMFC sensor system has been reported to be capable of determining isoflurane vapors from sweat below 40 ppm, with a sensitivity reaching 0.038 nA ppm −1 cm −2 , 26 and when using principal component regression analysis exhibited improved reliability ∼81% over a linear regression analysis during anesthesia (onset, maintenance, and emergence). The developed isoflurane fuel cell sensor is easy to fabricate, low in cost, and user-friendly. Analysis of the 8,032 data points in the study showed signal overlap in the nano-ampere range suggesting that external factors such as humidity could have influenced the signal's quality and require finetuning of the device, optimum anatomic placement, patient characteristics, and further validation of large clinical trials. It was also suggested that a baseline drift or transient fouling affecting the reaction rate in the electrodes could contribute to the signal overlap. This signal overlap issue was addressed by employing the PCR model in place of the Linear Regression model over the same data set, significantly reducing the deviation percentage. 26 An original configuration of a fuel cell sensor for anesthesia is an electrochemical DNA-based biosensor in a dual-chamber fuel cell system. 32 As displayed in Fig. 11 the anode and cathode reactions are separated into two chambers. The anode is externally wired to the cathode with a copper wire. Notice the graphite electrode of the cell is surface modified with gold nanoparticles, improving the charge transfer and demonstrating a significant decrease in electrochemical impedance.
Anodic oxidation reaction: Cathodic reduction reaction: Ethanol sensing.-Alcohol intoxication is the primary cause of road accidents in the U.S. and worldwide. The National Highway Traffic Safety Administration (NHTSA) data shows that 18.5 million people over 18 years old suffer from alcoholism. Moreover, car accidents related to driving under the influence (DUI) are the cause of death for nearly 11,000 people each year; and economic losses upward of $132 billion on average each year in the U.S. Existing alcohol breathalyzers cannot be used for continuous monitoring and have other limitations such as signal interference due to humidity, stability for continuous measurement, and working lifetime. 42 PEMFC sensors have been developed as a non-invasive, real-time, and continuous sensing method with accuracy and precision even for zero-tolerance alcohol consumption.
The reiterate theory of PEMFC sensing is based on the electrochemical redox reaction. Upon the intake of VOCs, excess gas  molecules are excreted from human skin where the sensor interfaces with the VOCs. Catalyst material determines the electrochemical surface-active area or sensing upper limit in MEA. The reaction rate depends on the relative humidity R.H. and protonic conductivity within the membrane. [43][44][45] The working principle of PEMFC-based transcutaneous ethanol sensing (left) along with the sensor prototype design (right) in Fig. 12. The ethanol oxidation and reduction occur at the two electrodes of PEMFC separately. After the oxidation reaction, acetaldehyde (CH 3 CHO) and acetic acid (CH 3 COOH) are produced as the intermediate products at the anode. The protons pass through the membrane across the electrodes while the electrons are forced to the external circuit. The stoichiometric equations in 11-13 show the resultant product of the electrochemical reactions. The number of protons and electrons is dependent on the concentration of ethanol. Thus, the current output is used as the fuel cell sensor signal. It is under the amperometry category in the electrochemical sensor classification. F.C. detection for ethanol is reported to be 0.047 nA ppm −1 , and the detection limit is 1 ppm. 9  Fuel Cell breathalyzers function similarly to the transdermal ethanol sensing explained above. Shown in Fig. 13b is the diagram of the breath flowing into the inlet, the chemical redox reaction occurring at the interface of the membrane, the electron moving through the external circuitry, and subsequently displayed in Fig. 13c. and the final reduction reaction occurs at the opposite side of the membrane. Breathalyzers do not offer the continuous monitoring and wearability that transdermal sensing provides. So, employing fuel cell technology optimized with modern techniques of machine learning to obtain ultra-sensitive and ultra-selective wearable ethanol sensors is ideal. 46 Acetone sensing.-People diagnosed with hyperglycemia must constantly monitor themselves to protect their well-being. Often, by the time they even notice, they are already suffering symptoms and need to act to restabilize their vitals. A better solution would be continuous monitoring for preventative maintenance, like the concept applied to machinery in the industry 4.0 trend. Acetone can be drawn as a signature for monitoring the vitals in hyperglycemic individuals. The increase of acetone in human sweat or breath is indicative of type-Ⅰ diabetes. It can evolve into "ketoacidosis"-a terrible hyperglycemic situation. Fuel cells designed and optimized for acetone can be the solution. Employing a fuel-cell sensor to realize the continuous realtime monitoring of the biomarker of acetone can potentially keep people aware of their health status and take instant (preventative) actions. A wearable device has been developed for acetone selectivity with a detection limit reported at 0.5 ppm to 4 ppm. 34,47 Glucose sensing.-Like the previous example, many people need to constantly monitor their glucose levels to maintain their Figure 11. A digital multimeter (DMM) read out of a fuel cell sensor system with an anodic/cathodic dual-chamber determines the anesthesia drug ketamine. 32 Figure 12. The working principle of the Transdermal PEMFC sensor for ethanol from human skin (left) and the watch-style prototype design (right). 43 optimal health. Current methods are often expensive and invasive. Glucose detection and quantification via electrons have been a method for many years. Glucose detection and quantification via electrons have happened for years. The discovery of the first enzyme electrode was by Clark and Lyons in 1962. 48 The outcome of tight glycemic control for managing glucose levels in diabetic patients is essential. Real-time constant glucose-sensing can significantly decrease the possibilities of diabetes-related diseases reminding people with diabetes to continue a healthy life while obviating the expensive and lethal late-stage diabetic complications. 48 Due to the importance of sensing glucose in diabetics, most glucose research concentrates on the glucose concentration in the blood. Currently, there are two ways of detecting glucose. There is the Enzymatic glucose detector and the nonenzymatic glucose detector. The fuel cell sensor can be used for the continuous real-time detection of glucose if the oxidation/reduction reaction and the catalyst are addressed. Here, we will discuss the mechanism and reactions of current glucose sensors.
The enzymatic glucose sensor is the one that is more commonly used. Different types of enzymes can be used for glucose sensing (Fig. 14). The most used enzymes are glucose oxidase (GO x ) and glucose dehydrogenase (GDH). For the two enzymes previously mentioned, GO x is the one that is used the most. Numerous glucosesensing devices use it due to the excellent selectivity of the glucose. It can tolerate fluctuating pH values and temperatures. Conversely, GO x loses most of its activity below pH-2 or above pH-8 and can be irrevocably destroyed at temperatures higher than 40°C. 49,50 Exposure to unstable humidity can negatively affect the GO x -based sensors and depends much on oxygen. Glucose dehydrogenase is used with cofactors such as pyrroloquinoline quinone (PQQ) and flavin adenine dinucleotide (FAD). This enzyme and coenzyme have a better performance than that of GO x . It is widely applied in blood glucose sensing at present. 49 Nonenzymatic glucose sensing is an alternative to enzymatic sensors; a fuel cell sensor for glucose is a non-enzymatic process (Figs. 15 & 16). Due to enzymes' nature, Enzymatic sensors have limitations such as pH, temperature, and oxygen dependence. Nonenzymatic glucose sensing devices use direct electron catalytic glucose oxidation on the electrochemically active sites of the electrode. 48 Therefore, it does not have the same limitations as the enzyme sensors. Different oxidants can be used for non-enzymatic sensors. From metallic redox to nanostructures have been used as  non-enzymatic detectors. As electrocatalytic processes take place through the adsorption process, the analyte molecules are adsorbed into the active sites of the electrode. Usually, the electrodes used for fuel cell sensors are platinum and gold, but metal alloys can also be used. The metallic redox works by multiple factors: favorable electronic states of the redox center, unfilled d-orbitals at transition metal centers, and abundant defects in non-metal-based catalysts. After Absorbing targeted molecules, it goes to chemical bond breaking and thus forming intermediates. Since the oxidation state of the redox center is different, the interaction between the product and electrode is less intense, resulting in the desorption of the product from the material surface. The adsorption-desorption of target molecules at the surface is the chemisorption process (Fig. 15). 49

Advanced Materials for PEMFC-TAMS
State of the art of MEA materials.-The Membrane Electrode Assembly (MEA) is the core of Fuel Cell technology. The membrane must be ionically conductive and electrically insulative; while the electrodes and catalyst loadings are critical for oxidation and reduction reactions. Each detail must be evaluated, from the dispersive solutions used to integrate the electrode onto the bare membrane, fabrication methods and parameters involved, down to the end-of-life performance. MEAs must be carefully designed considering performance, durability, selectivity, cost, and more. The materials in PEMFC sensors are critical to the sensing functionality and should be tailored for specific applications. To ultimately improve the performance and durability, and develop the most cost-effective sensors, we need to comprehend, document, and share our investigations of the countless variables.
A crucial component to facilitate electrochemical reactions in PEMFC is the catalyst layer employing electroactive materials. The most widely used set among these electroactive materials is the platinum group metals (PGM). 52 Despite the now-available commercial deployment of fuel cells, efforts to reduce their cost and durability are still a relevant topic of discussion. While PGM catalysts have excellent electrochemical performance and efforts have been made to reduce the amount of material required in PEM fuel cell systems, they still represent the highest cost in these energy conversion systems. Therefore, the work to develop PGM-free catalysts that compete with their precious metal counterparts has been performed since the 1960s. 53 Recently, advances in the PGM-free catalyst field have shown that significant improvements in performance (current density) and durability can be achieved by tuning active site density in iron-based catalysts. [54][55][56] Implementation of PGM-free catalysts in commercial systems has been a topic of interest in recent years due to the potential cost savings this could entail (>40%). 57 Despite the performance and durability increase achieved in recent years; these parameters are still not up to par with most industry requirements where PGM-based fuel cells are used. 58 While PGM-free-based fuel cell systems could be used for low-power, low-longevity applications, significant improvements must be made for them to become viable at a large scale and compete with PGM-based systems.
Despite considerable research efforts dedicated to developing novel catalyst materials for PEMFCs, such as the ones listed above, these devices require their membranes to possess fundamental properties to perform at an ideal level. Crucial properties that membrane materials must exhibit are mechanical and thermal stability at their operating conditions, chemical stability to avoid degradation due to H.O.· and HOO· radical formation, and great ionic conductivity properties while preventing electron transport hydrogen-oxygen gas crossover. 59 The most widely used membrane material that operates below 100°C is Nafion ® . Nafion ® is a perfluoro sulfonic acid-based polymer exhibiting excellent protonic conductivity and chemical resistance. However, this high proton conductivity is only exhibited when the material is hydrated and therefore can only be used in a limited temperature range. [60][61][62][63] Due to the limited conditions in which Nafion ® can operate, efforts to develop new membrane materials have grown in the past. The most studied materials for novel polymer membranes are polybenzimidazole, polyetheretherketone, and polysulfones. [60][61][62][63][64][65] In addition to these research efforts into suitable polymers for different operating conditions, composites with hygroscopic ceramic reinforcements have also been investigated. The inorganic materials used for this application are TiO 2 , TiSiO 2 , and ZrO 2 . 63,66 In the membrane, the hydrophilic group controls the dynamics of proton transport. The transfer mechanism is known as the cluster channel model. 67 The sulfonated acid groups become water-filled clusters where the ions move through. It is indicative that the Figure 15. Chemisorption and oxidation of glucose in fuel cell sensors. 50,51 protonic transport in the Nafion ® membrane is highly dependent on the hydration level. The reaction rate and performance are determined by R.H. and the temperatures in the environment. For fuel cell sensors, this will decrease the stability and reliability of the sensing function. The current output is not determined only by the concentration of the analyte but also by the humidity. Therefore, eliminating the interference of humidity in a dynamic environment is critical in real applications. Developing a membrane that functions independently of humidity is one of the potential solutions and one of the most challenging.
Pre-concentration in MEA design.-Some research has been done to address the insufficiency or improve the performance in PEMFC sensors regarding the MEA configuration or specific components, e.g., electrode designs, catalyst, membrane, and MEA design. The two-electrode and three-electrode setups with Nafion ® as PEM are compared, and the three-electrode electrochemical system shows higher stability of the two. The catalysts, including copper, stainless steel, and nickel, have also been compared. Nickel has the most excellent catalysis behavior for ethanol oxidation and oxygen reduction. Its current response is 300 times and 3 times better than stainless steel and copper, respectively. 45 Incorporating the pre-concentration process is in the interest of improving the sensing reliability of PEMFC for VOC. A threedimensional structure for heavy molecular loading has been obtained with multi-walled carbon nanotubes and gold nanoparticles for an electrochemical biosensor. The enhanced electrochemical response has confirmed the strategy's success. 68 Nanomaterials play a pivotal role in electrochemical biosensing due to their abilities to build bioanalytical platforms and improve the transduction of biorecognition events. Based on the theory and the preliminary experimental validation, the preconcentration with nanomaterials is expected to improve the sensitivity and reliability of the fuel cell sensor. The fuel cell hardcore theory, fabrication, and modification are central to sensing applications.

PEMFC-TAMS Technology
Recently the micro-PEMFC sensor has proven a promising technology for VOC monitoring. The unique advantages are simplicity, fast response, broad sensing range, real-time, and portability. [69][70][71][72][73] The majority of the PEMFC sensors use Pt as the catalyst and Nafion ® as the membrane. The use of alternative catalysts and natural membranes such as eggshell membranes have also been explored for sensing purposes. 74 Given sensing application requirements, a successful device for continuous monitoring needs to be constructed by the following steps, (i) identifying and developing bulk metal and nanostructured metal catalysts, (ii) implementing and integrating them into the anodes and cathodes of the fuel cell setup, (iii) and optimizing electrochemical performance to determine the detection limit, sensitivity, stability, and specificity.
Case demonstration.-The team at Florida International University has been working on the PEMFC sensor development 45 in collaboration with Los Alamos National Lab. Based on the results obtained in these studies, a wearable device with an ultra-low-power MSP430 microcontroller together with IoT (Internet of Things) has been constructed (Fig. 17). A wearable "fit bit" type platform has been built to center the fuel cell. The miniaturized wearable electronics include electrochemical gas sensing integrated circuit LMP91000, a custom-designed printed circuit board, Bluetooth enabled, with an accompanying android application, and a rechargeable lithium-ion battery. Bluetooth is for the wireless transmission of sensor data (nano-ampere range current) to a smartphone and is attached to the wrist or forehead of people in need. A pipeline for efficient transport of individuals in unmanned drones based on the wearable PEMFC micro-sensor has been proposed. 45 Baseline readings for calibration of the device need to be obtained before using sensing (Fig. 18). Some of the calibration baseline for ethanal and isoflurane has demonstrated the linear relations between the current from the fuel cell and the analyte concentration. 32,44 In all cases, anesthesia was maintained with isoflurane (0.5%-2%). Overdose of isoflurane inhalation for patients and staff during surgery in the operating room has severe consequences for people's health. Assessing the sensitivity and accuracy of isoflurane concentration needs to be done via high-cost and sophisticated infrared (I.R.) spectroscopy continuously in the operating theater. The PEMFC-TAMS technology will potentially function as an anesthesia sensor at a low-cost, minimized platform, on-site, continuously, and in real-time. In addition, the fuel cell has a promising future in ethanol sensing to increase the awareness of alcohol content in blood in social drinking, reduce drunk driving and help law enforcement. It can be potentially used as an alternative device for sensing continuously anytime and anywhere (Fig. 19).
PEMFC-TAMS calibration.-Surgical procedures require administering anesthetic agents in a 0.5 to 2% concentration. These agents are combined with oxygen, air, or other sources and received by the patient through inhalation to sustain sedation. 75 The common anesthetic used in surgery is isoflurane. An adequate isoflurane dose is critical for the procedure since a variation in concentration on the subject can cause an overdose or awareness during surgery. 76,77 The equipment is responsible for analyzing anesthetics' concentration inhaled and exhaled and controlling the dose delivered to the subject throughout the surgery, using vaporizers and I.R. sensors. 78,79 These modern machines are very complex, require a golden stage to operate correctly, and are of high cost, limiting their deployment in low-resource areas, increasing the risk of poorly anesthetic administration and adverse health outcomes. An accessible, low-cost anesthetic sensor could improve surgical procedures' safety and positive health outcomes. 26 PEMFC sensor is studied for the selective detection of compounds for point-of-care (POC) diagnosis of the physiological status of people. A PEMFC is an electrical transducer converting chemicals to an electrical signal. A chemical reaction occurs at its electrodes, where the concerned VOC biomarkers (e.g., isoflurane, ethanol) are oxidized. Calibration requires a well-controlled system for biosensing applications.
The implementation consists of utilizing the simplest form of a fuel cell sensor, PEM sandwiched between two metal electrodes to measure the isoflurane excreted through the subject's skin. Conditioning and calibrating the fuel cell sensor to receive a chemical component, such as isoflurane, is crucial for acquiring a reliable and accurate device. Figure 20 represents calibration setups.
One illustrates a calibration setup that has been implemented, and the other illustrates a setup that could allow for more accurate calibration of the device.
The setup (Fig. 20) allows calibrations of the fuel cell for ethanol contaminants. 26 By changing ethanol to another type of VOC, isoflurane, Dr. Jalal's work can calibrate the fuel cell sensor to detect different anesthetic concentrations. 80 Even though the setup allows for this type of calibration, it presents some issues that could hinder the data gathered throughout the experiment. The input gas pressure is a crucial parameter that affects the fuel cell measurement. With the represented setup proposed by the authors, gas flow can be controlled but not the pressure, thus affecting the fuel cell functionality and the data collected. Not controlling the pressure also leaves the system vulnerable to leaks, hindering the data's reliability. Another poorly controlled parameter in this setup is the input gas concentration, which prevents the accurate translation of voltage produced by the fuel cell to contaminant concentration. Implementing the proposed setup can potentially solve all the inaccuracy sources. The system pressure is maintained constant because it is built with metal tubes and connections, thus allowing the control of the gas flow and the system pressure. Also, the setup provides a better input gas concentration mechanism than utilizing a hot plate with an isoflurane container and allows the control of liquid quantity, concentration, and evaporation rate (Fig. 20). Furthermore, turning the gas valve on will carry the desired concentration of isoflurane into the fuel cell, thus providing more reliable data and a more accurate calibration experiment.

PEMFC-TAMS Modeling
Modeling fuel cell sensors provides essential tools for predicting the device's function for the sensing materials with enhanced sensitivity. Many models have been developed. Some are physicchemical, and some are based on integrated equivalent circuit emphasis considering an electrochemistry system. This section describes some theoretical analyses used for many PEMFC models, providing a deep insight into the working mechanism and potential prediction.
Generating accurate equivalent circuit models (ECMs) can establish conventional methods of characterizing behavior. 81 Equivalent circuit modeling translates the behaviors of complicated systems, such as any electrochemical system, into simplified electrical circuits. Analyzing the system solely in terms of its electrical properties is a way to characterize a system comprehensively. Models consist of lumped elements and frequency-dependent elements. "The lumped elements are resistance, capacitance, and inductance, and the frequency-dependent elements compromise of the Warburg element, constant-phase element, bounded frequencydependent element, bounded constant-phase element, and unloaded phase element." 82 The methodology of using electrical components can establish a baseline representation of a system which can then be adjusted appropriately to fit the behavior of other similar systems. It can even be improved upon when further system analysis yields a new understanding of the system. An excellent example is the model creation began by comparing actual data of their PEMFC with the Randles cell in the Warburg finite length diffusion element model. 83 The data did not correlate exactly, especially in the high-frequency ranges. The paper outlined how a constant phase element (CPE) could improve the model to fit the data better. What is imperative to note is that the paper also stated the critical importance of verifying that this additional element correlates with some physical property of the system. If there is no confirmation, adding an element to fit the data creates an inaccurate system model. The data's detailed statistical analysis with the new model proved that this element corresponds with physical property. Additionally, the study verified that the new model fits the data within the "threshold in which it would be useless to refine further the model." 83 Numerous baseline models exist for PEMFCs. A comparison of several of these models, including Randle's model, is in Table II. Once a model is adequately constructed, it is a great tool for predicting a system's behaviors. For example, if the electrochemical system becomes a component within a more extensive system, having a precise model of the electrochemical components can lead to better predictions for the overall behavior of the more extensive system. In the more intricate models, we can see a representation of more specific system components. Certain models have generic applications, while others have a specific scope.
Advanced-data analytics for PEMFC-TAMS.-Data generated from a lab shows clear relationships between input and output. However, under real life applications there are multiple variables that influence a system. Normally calculating all these numerous data points to define a relationship can be time consuming and problematic. With the use of machine learning (ML) this becomes a lot simpler. With a minimal amount of code, a relationship between one input and one output systems can be defined, as if it were a normal polynomial function, and with more sophisticated ML methods you can define relationships between 1-5 different inputs with varying outputs.
Design optimization of micro/macro-fuel cell sensors for future application in advanced data analysis should deploy a platform that allows fast sampling speed. To provide sufficient data for deep learning (DL) methods, current fuel cell sensor research and developments need to improve stability, durability, and reliability with pre-autocalibration procedures for drift correction. Moreover, the standard control of calibration and measurement plays a significant role, even though this varies with the target analyte and the active material in the sensor platform.
Sensing becomes a challenge in a multivariate environment. R. H., Temp, and the interference of other VOCs with the sensor's functionality give unpredictable outcomes. As a result, precisely correct VOC(s) calculation and calibration are the main difficulties in sensing applications. A statistical model can be derived to address these difficulties in calibrating fuel cell sensors by deploying multivariate data analysis, such as the "Principal Component Regression (PCR)" advanced data analysis. PCR can link broad data sets and give reliable data fitting between a known and an Figure 19. PEMFC micro-sensor as an alternative device for VOCs monitoring. unknown data set. PCR can increase the accuracy in complex environments compared with linear (univariate) regression models. 26 Several challenges affect the fuel cell sensing accuracy, sensitivity, and practicability. One of these issues relates to multimodal sensor data and how best to analyze the data to gain meaningful process reliability information. In many environments, there is a lack of well-controlled enclosed facilities from the surroundings. The uncontrolled dynamics under different conditions influence the sensitive devices' output, resulting in significant variations in readings over time during clinical trials. This project deals with these issues by focusing on the calibration aspect: (i) Multimodal technique to eliminate interference; The multimodal electrical method, including open circuit potential (OCP) and amperometry measurements, are incorporated and applied to stabilize the detection of the analyte where the disturbance exists. The sensor is interfaced with a miniaturized electrochemical station, with a microcontroller to proceed with an auto-calibration algorithm tailored to increase the specific analytic isoflurane sensing reliability. Multimodal electrochemical techniques can eliminate the signal of the background noise. (ii) Multivariate calibration approach. Principal component regression (PCR) will be employed to establish the link of response variables to predictor variables to improve the calibration. Larger data sets can be highly correlated, and the fitting between known and unknown data can also be enhanced with PCR. 26 The PEMFC sensor problem (interference of humidity, intricate calibration issues, etc.) can be easily identified and solved using artificial intelligence (A.I.) data analytics. The data streams from the sensors can be evaluated using three potential machine learning (ML) methods, namely neural networks, support vector machines, and multi-gene genetic programming (MGGP).  Humidity's (R.H.) influence on the output was recorded as a time series model at intervals of (time-every minute/hour/sec) and evaluated using the three A.I. models. These models can process minute changes in the data and have proved practical for making minor sensor adjustments for calibration purposes. The data is classified into 70% for training, 15% for validation, and 15% for the testing set at random. For the PEMFC sensor application, the experiments can be separated into two parts: 1. To study the impact of various R.H. on the sensor, the data will be recorded at several temperatures at intervals of 5 degrees Celsius; 2. To calibrate/ measure the concentration of the analyte, data from various sensors such as alcohol, anesthesia, and other volatile organic compounds to quantify the concentration influence on PEMFC sensor performance.
ML methods.-1. Neural networks -N.N. can master fast and be employed as a modeling tool for many applications. It comprises three layers: an input layer taking the input variables, a hidden layer consisting of neurons, and an output layer number of output variables as the neurons. 2. Support vector regression-Initially based on support vector machines (SVM). It has been applied to address classification issues. SVM used for regression is recognized as SVR, which is not void of statistical assumptions such as model structure assumption, and error dependency, and formulates models based on the collected data. The framework of SVR is derived from the structural risk minimization (SRM) principle, which modifies the empirical risk minimization (ERM) principle and minimizes an upper bound on the expected risk. 3. Multi-gene genetic programming (MGGP) provides a structural optimization method where tree structures present the solutions; this works very similarly to a Random Forest model where each model working in the evolutionary process consists of a few sets of trees combined in the MGGP model. This method is weighted linear incorporation of output values from the number of trees.

Power Awareness in Sensing Design
Uniquely for portable sensing applications rather than classic stationery in lab characterization systems, it is critical to consider the computing features and techniques necessary for accurate sensor reading with minimal power consumption. It is impractical for wearables and other portable devices to be charged while in use; therefore, power management is important to avoid depleting the battery at an unacceptable rate. Advanced features such as continuous monitoring naturally consume more energy than their noncontinuous monitoring counterparts; standard tasks performed for continuous monitoring include sensing, data analysis, and wireless data transmission. Understanding which task is the largest source of energy consumption will be beneficial in implementing the right protocols to minimize energy consumption without sacrificing sensing capabilities. Ultimately the concept of trade-offs and finding the perfect balance between energy consumption and necessary computing power for accuracy will enable device optimization. 27 Intricate networks such as human movement sensing networks are beginning to trend in research and could soon progress towards applications such as full body-medical diagnostic methods. With the advancement of the wearable sensor, a full-body array of sensors could even be employed to monitor the health of soldiers stationed in remote areas. Like the concept of the trade-off that is employed for feature selection for a single sensor device with a power-aware and accuracy as limiting values trade-offs of which sensors are necessary for correct classification.
As reported by Ghasemzadeh, the case study investigating human movements determined that not all sensors were necessary for correct classification. By employing algorithms to determine which sensor to eliminate due to inefficient energy consumption, they could eliminate or turn certain sensors for different movements. 84 Understanding the power consumption that coincides with machine learning classification, trade-offs for optimized sensing and power consumptions, continuous monitoring of power consumption, and circuit design for power and thermal awareness will substantially contribute to sensor device design.

Conclusions
This paper has reviewed the recent research and development of PEMFC electrochemical sensors in the practical application of noninvasive continuous real-time monitoring of VOCs in humans. To successfully design a market-deployable wearable sensor it must be aesthetically pleasing, low-cost, low-power, non-invasive, and continuous monitoring without sacrificing selectivity, sensitivity, and stability. Thus, highlighting the benefit of fuel cell (FC) wearables for point-of-care sensing technology. The redox reactions of fuel cell sensors for different VOCs, including ethanol and isoflurane, are compared. The isoflurane sensor is reported to have a detection limit of 40 ppm with a sensitivity of 0.038 nA ppm −1 . Principal component regression improved reliability by ∼81% compared to the linear regression analysis. An ethanol sensor is reported to have a limit of detection at 1 ppm with a sensitivity of 0.047 nA/ppm. These values demonstrate that the PEMFC-TAMS technology is promising for wearable sensor devices. This work classified and summarized the working principles, advanced materials, and theoretical modeling for different PEMFC sensors. We have also summarized the AI/ML algorithms employed in the sensors' data streams and introduced the concept of power awareness for sensor device/network design. These techniques allow for a fast sampling rate to retain the value of the fast response time in PEMFC-TAMS sensors. This creative fuel cellbased sensor can bring new ideas to existing electrochemical or solid-state biosensing studies, point-of-care, and disease diagnostics applications at the early stage; it also broadens the applications of fuel cell technology for energy generation and bridges the interdisciplinary subjects.