Flexible, Fibrous, and Rigid Chemiresistive VOC Sensors with Nanoparticle‐Structured Interfaces

As one of the noninvasive screening and diagnostic tools for human breath monitoring of various diseases, chemiresistive devices with nanomaterials as the sensing interfaces for detecting volatile organic compounds (VOCs) have attracted increasing interests. A key challenge for the practical applications is an effective integration of all components in a system level. By integrating with the system components, it provides reliable and rapid results as a fast‐screening method for healthcare, safety, and environmental monitoring. This paper highlights some of the latest developments in chemiresistive sensors designed for the detection of VOCs and human breaths. It begins with a brief introduction to the fundamental principles of chemiresistive sensors with nanoparticle‐structured sensing interfaces. This is followed by a discussion of the recent fabrication methods, with an emphasis on nanostructured materials. Some of the recent examples will be highlighted in terms of recent innovative approaches to sensor applications and system integrations. Challenges and opportunities will also be discussed for the advancement and refinement of the chemiresistive sensor technologies in breath screening and monitoring of diseases.


Introduction
[3][4][5] The human breath comprises a complex mixture of various gases, such as nitrogen, oxygen, carbon dioxide, water vapor, and over 1000 other compounds.The rich and diverse composition of breath gases makes it a valuable source of information for understanding various physiological and pathological processes in the human body, offering potential insights into the detection and monitoring of different diseases and health conditions.Certain VOCs in the breath have been recognized as potential biomarkers for specific diseases. [1,6]he detection of such VOCs provides diagnostic information.However, detecting VOCs faces significant challenges, primarily because many of these compounds are nonreactive and exist in low concentrations, usually close to or even below the detection limits of the detection techniques.Among the most robust and extensively employed techniques for detecting VOCs is gas chromatography coupled with mass spectrometry (GC-MS). [7]GC-MS provides comprehensive analysis, quantification, and identification of a broad spectrum of compounds, making it well-suited for VOC detection.However, this method also comes with some evident drawbacks.The requirement for relatively bulky, expensive, powerand time-consuming equipment, along with the need for trained personnel, are notable disadvantages of GC-MS.These factors limit its accessibility and practicality in certain settings, prompting the exploration of alternative approaches for VOC detection.Considerable attention has been devoted to VOC sensors, resulting in the availability of various sensor types in both commercial markets and research laboratories. [8,9]However, some of these sensors have limitations in terms of sensitivity, selectivity, and the requirements for effectively detecting the diverse and unpredictable combinations of VOCs presented in the samples.An ideal VOC sensor should display high sensitivity, selectivity, and reliability, and exhibit rapid response time.In addition, the low-cost fabrication is also important for practical applications. [10,11]These criteria have spurred significant efforts in the pursuit of structurally tunable VOC sensing interfaces and versatile methods over the past few decades.48][49][50][51] Some of the important aspects of the wearable sensors and applications have been highlighted by several outstanding and comprehensive reviews. [19,20,52][55] Part of this approach is illustrated in Figure 1 in terms of the integration of chemiresistor sensor arrays derived from nanoparticle-structured sensing interfaces with sensor electronics and artificial intelligence (AI) in detection of human breath VOC biomarkers for screening and diagnostics.In this report, we highlight some of the examples, along with the advancement in chemiresistive sensors for breath VOCs detection.It starts with a brief introductio to the fundamental principles of chemiresistive sensors with nanoparticle-structured sensing interfaces.This is followed by a discussion of the recent fabrication methods, with an emphasis on nanostructured materials, and selected examples demonstrating the promising appli-cations of the chemiresistor sensor system for detection of human breath VOC biomarkers for screening and diagnostics.

Response Characteristics
The sensing characteristics of a chemiresistive sensor with nanoparticle-structured interfaces are primarily determined by the interparticle molecular interactions.These interactions significantly influence the collective properties of the nanoparticleligand ensemble, including electrical and optical properties.The adsorption of gases can cause changes in the intermolecular interactions, resulting in alterations to the electrical properties of the chemiresistive sensors.The correlation between electrical conductivity and nanostructural parameters, such as particle core radius (r), interparticle distance (), and dielectric constant of the interparticle medium (), is a crucial factor when utilizing nanostructured thin film materials to create chemiresistive sensing arrays.These parameters directly influence the activation energy in thermally activated conduction paths, consequently playing a significant role in amplifying the electrical signal in sensing applications.The overall electrical conductivity () can be described by a thermally activated conduction path: [56] (, T) =  0 exp(−) exp where e is the elementary charge = 1.6 × 10 −19 C,  0 is the vacuum dielectric constant = 8.854 × 10 −12 F m -1 , R is the Boltzmann constant = 1.38 × 10 −23 J K -1 ,  0 is a pre-factor, T is temperature,  is dielectric medium constant, and  is electron coupling constant.The activation energy (E a ) for activated electron tunneling depends on r,  and .Experimental results have confirmed the precise control of electrical conductivity concerning parameters such as , r, , and .This controllability has been demonstrated by self-assembled nanostructures obtained from gold nanoparticles with various sizes and molecular linkers with differing spacing. [56,57]The electrical conductivity of these materials can be finely tuned and adjusted by manipulating the specific parameters.
The electrical conductivity of the sensors may undergo changes when they are exposed to the target analytes.When a chemiresistive sensor is exposed to VOCs, the sensing film can swell or shrink which is responsible for the changes in the electric resistance.This sensing film, comprising both inorganic nanomaterials responsible for electrical conductivity and an organic film element facilitating VOC adsorption, [58] plays a crucial role in the sensor's response.Upon exposure to VOCs, the organic film element interacts with these compounds, leading to either swelling or shrinking of the sensing materials which induces changes of the volume in the nanomaterial film.These volume changes directly impact the electrical resistance of the sensing film.Consequently, the sensor exhibits a measurable change in its electrical resistance, and this change is used to detect and quantify the presence of specific VOCs.
The relative sensor response, ΔR/R i , can be assessed by a theoretical model in terms of thermally-activated electrical .Adapted with permission. [54]Copyright 2023, ACS Publication.conductivity, [54] which is expressed as the relative resistance change In this equation, in addition to the regular constants (e is the elementary charge ), T is temperature,  is the electron coupling term, r and  represent particle radius and interparticle spacing respectively, [59,60]  sw represent the dielectric of the analyte swollen ligand matrix,  film is the dielectric of the sensing film, and Δ is the change of the interparticle distance upon sorption of solvent.Equation 2 describes how the electrical conductivity changes when VOCs are adsorbed in the film.This change is expressed in terms of the relative change in resistance ΔR/R i , following the principles of thermally-activated conduction theory.In this context, two essential parameters are introduced: Δ (change in interparticle distance) and  sw (dielectric constant of the medium), which play crucial roles upon VOC adsorption in the sensing film.Both Δ and  sw are influenced by various factors, such as particle size, interparticle distance, and the concentration of the vapor (C v ).
Theoretical simulations have been conducted to predict the sensor responses towards different lung cancer biomarker VOCs.The simulated responses for a 9-sensor array to different panels of VOCs associated with lung cancer were analyzed by principle component analysis, as shown in Figure 2, [54] showing effective separation of the different VOCs.To validate the simulation, the simulated results were compared with experimental data.The principal component analysis (PCA) plots from both simulated and experimental data clearly indicated the separation of the major VOCs, demonstrating the sensor array's capability in achieving high recognition selectivity.Importantly, there is a clear agreement between the simulation and experimental results, further supporting the validity and reliability of the simulation approach.

Sensor Fabrication
A chemiresistive sensor consists of two crucial components: the sensitive sensing film and the electrical signal transduction electrodes.One of the important approaches to the sensitive sensing film with the desired sensing sensitivity is to manipulate the morphologies and structures of nanoparticles and ligands in the nanostructured thin films.Various fabrication methods can be employed to create the sensing thin films on chemiresistive sensors, such as drop casting, layerby-layer assembly, molecularly-mediated self-assembly, microcontact stamping, and ink printing. [17,61]Each of these pathways offers unique advantages and can be tailored to meet specific sensor requirements.In the past, several approaches have Figure 3. Illustration of the fabrication of chemiresistive sensors derived from Au NPs on flexible substrates printed with microelectrodes via surface/interfacial linkages or interactions of alkanethiols and functionalized alkyl thiols to define the interparticle spatial properties.Adapted with permission. [62]Copyright 2021, Elsevier.
been employed that start with the synthesis of molecularlycapped metal nanoparticles, with a focus on coinage metals like Au, Ag, Cu, and their nanoalloys.For these molecularly capped nanoparticles and molecularly-mediated self-assembly techniques, the surface and interfacial chemistry play a crucial role, as demonstrated in ligand exchange reactions and ligand cross-interactions between nanoparticles, capping molecules, and mediator molecules.These processes involve creating interparticle linkages and the assembly of the nanoparticles on the substrate, which play a significant role in determining the structural integrity and properties of the resulting chemiresistive sensor.In Figure 3, selected examples are included, illustrating the capabilities of manipulating surface/interface chemistry using alkanethiols, functionalized alkyl thiols, and gold nanoparticles.This manipulation allows for the precise definition of interparticle spatial properties within ensembles or thin films.These ensembles or thin films can be effectively formed on different substrates, including flexible substrates like PET, PI, papers, and other materials, with interdigitated microelectrodes printed on them.Consequently, these sensor arrays can be utilized as chemiresistive sensors in a wide range of potential applications.
In the traditional fabrication process of electrical signal transduction electrodes, photolithographic patterning is commonly used.However, this approach comes with certain limitations, including high costs, material limitations, and limited scalability in manufacturing.To overcome these limitations, an alternative method has gained popularity -printing conductive inks with low-temperature and roll-to-roll processing conditions.This approach has become increasingly appealing as it offers the advantage of fabricating electrodes on various substrates including paper, providing greater flexibility and cost-effectiveness in the production process.Recent studies have demonstrated various approaches [63,64] that emphasize the use of nanomaterials as printable inks.One approach involves using AuCu alloy nanoparticles for fabricating chemiresistive sensors.These approaches provide further evidence of the growing interest in utilizing such nanoparticles as versatile building blocks for sensor applications.An exemplary case involves the application of pulsed laser for nanoparticles sintering on flexible substrates. [63]In Figure 4A, the process of pulsed laser sintering is depicted, illustrating how it offers an efficient method to transform printed AuCu nanoalloy with adjustable size and composition into a conductive pattern on a flexible substrate.Additionally, this technique enables precise control over the thermal penetration, resulting in a tailored conductive footprint on the flexible surface.The nanoalloy exhibits improved air stability compared to pure Cu NPs and reduced melting point compared to bulk alloy counterparts, owing to its nanoscale alloying.In the process, a focused pulsed laser beam is precisely scanned over the substrate printed with the nanoink.This is done at specific speeds and power settings, resulting in the sintering of the nanoparticles.Any unsintered ink is subsequently dissolved as shown in Figure 4B.This approach allows for the creation of a conductive pattern with excellent precision on the flexible substrate.Another fascinating application highlights the room-temperature sintering of AuCu nanoparticles on fibrous and flexible substrates, [64] enabling the creation of cellulosic fibrous sensors in 3D dimensions.In this approach, microelectrodes are fabricated through surface-mediated interconnection (SMI) of nanoparticles, while the sensing film is embedded using surface-mediated assembly (SMA) of nanoparticles, as illustrated in Figure 4C.This innovative technique paves the way for the development of versatile and adaptable sensors with fibrous structures, opening up new possibilities for diverse applications.In contrast to conventional random interparticle necking pathways achieved through high-temperature sintering or cold welding, the SMI pathway presents distinct advantages, particularly when working with fibrous materials.Conventional methods often suffer from poor adhesion to fibrous substrates, limiting their efficacy.However, the SMI pathway overcomes this challenge by facilitating surface connection of atomic layers .Adapted with permission. [62]Copyright 2021, Elsevier.
among nanoparticles, leading to strong adhesion and potential patternability on fibrous materials.Moreover, the SMI pathway results in the development of metallic conductivity, both on the surface and through the thickness of the substrate, adding to its versatility and application potential.The integration of SMI-SMA coupling permits the precise printing of a microelectrode pattern on the fibrous substrate, followed by the subsequent printing of the sensing film, as depicted in Figure 4D.This innovative approach holds the potential to revolutionize the design and fabrication of 3D fibrous sensors, offering new possibilities and advancements in the field.
In recent studies, the approach is further expanded for creating a nano-filamented textile sensor platform [65] as shown in Figure 5.The key concept involves introducing both electrical conductivity and chemical functionality to an electrospun fiber network.By carefully controlling the porosity and hydrophilicity of the interlaced electrospun polymer fibers and nano filaments, a highly adaptable network with the desired sensing properties is achieved.The result is a flexible, low-cost, and disposable paper-like substrate format, offering immense potential for a wide range of practical applications.Findings from this work demonstrate that the response characteristics of chemiresistive sensors on the fibrous platform can be finely tuned by manipulating the size of dendronized gold nanoparticles and the size of the fibers within the fibrous microstructure.This approach provides the capability to introduce various functionalities in terms of conductive and sensitive filaments.Dendron-mediated gold nanoparticle assemblies on fibrous substrates with different diameters are explored.The introduction of dendrons imparts a combination of hydrophobic, hydrophilic, and dielectric ).Adapted with permission. [65]Copyright 2022, ACS Publication.microenvironments, allowing for precise control of intermolecular interactions at sites such as ─OH, ─(C═O)─, ─O─, and ─N─.Contrasting traditional 2D rigid sensing platforms with functionalized nanoparticles, these 3D fibrous platforms offer several advantages.Notably, they enable a higher density of sensing sites, resulting in enhanced sensitivity, further advancing the capabilities of chemiresistive sensors on fibrous substrates.
Another intriguing approach is recently demonstrated by Haick and co-workers for establishing a reliable method in fabricating chemiresistive sensor chips based on monolayer-capped metal nanoparticles (MCNPs), [12] as shown in Figure 6.The process ensures reproducibility across different devices and batches, offering consistent results and performance.In this study, the researchers have successfully addressed the issue of low reproducibility in chemiresistors.The foundation of this approach involved an optimization of the synthesis process for monolayer-capped gold nanoparticles (MCGNPs). [66]By meticulously controlling the synthesis temperature and the injection rate of the reducing agent, a precise manipulation of the core size and distribution of the MCGNPs were achieved.This level of control directly influenced the sensitivity and stability of the sensing film formed using these nanoparticles.Subsequently, to tackle the issue of uneven drying and the formation of coffee rings on microelectrode devices of deposition of the MCNP solution, a physical microbarrier around the drying droplet was introduced.This innovative approach effectively mitigated the undesired effects, ensuring a more uniform and controlled drying process.Functionally, eliminating the formation of coffee rings led to a more dependable response from the resulting chemiresistor devices.While minor device-to-device variations were expected and attributable to the precision of the analysis, the incorporation of the microbarrier ring undeniably resulted in a substantial reduction in the overall variability observed.

Chemiresistive Detection and Monitoring of VOCs
The nanostructured chemiresistive sensors have been increasingly utilized for detecting and monitoring VOCs, catering to disease detection vis human health monitoring.Some of the recent examples are listed in Table 1, featuring chemiresistive sensors for breath analysis or sensing in terms of the sensing nanomaterials and the detection targets and results.Different methods have been used for the formation of the sensing films.The thin films fabricated by surface assembly of nanoparticles (SAN) explore molecularly-mediated interparticle ligand exchange and crosslinking reactivities. [62]

Detection and Diagnostics of Lung and Gastric Cancers from Exhaled Breath
The nanosensors derived from gold nanoparticles with diverse organic ligands have been investigated in diagnostic tests for lung and gastric cancers by analyzing exhaled breath samples. [69]s previously described, the sensors were fabricated using the micro-barrier ring method. [12]The sensors are derived from layers of gold nanoparticles functionalized with 13 distinct organic ligands and prepared in either manual drop casting or printed Figure 6.Illustration of the fabrication process of MCGNP-based sensing chips. [12]A) Schematic depiction of chips with eight MCGNP-based sensors with microbarrier (the dark edges on the outer side of the electrodes are the ring-like barriers).B) Inkjet printing about interdigital electrodes.C) MCGNP on the surface after drying.Adapted with permission. [12]Copyright 2021, Springer Nature.chips leading to the creation of 26 unique sensors.The study collected and analyzed a comprehensive dataset comprising 545 breath samples obtained from 426 adult participants using the gold nanoparticle-based sensors mentioned above.This diverse group consisted of 158 lung cancer patients, 115 gastric cancer patients, and 153 healthy volunteers.The dataset was segregated into three distinct binary comparisons: lung cancer versus control, gastric cancer versus control, and gastric cancer versus lung cancer.For the analysis, 70% of the data was utilized for training purposes, while the remaining 30% was reserved for testing.Statistical methods include discriminant factor analysis (DFA), and receiver operating characteristic (ROC) analysis were applied to the sensor responses to derive meaningful patterns and evaluate the diagnostic performance of the sensor array effectively.
In the training set, the classification model designed demonstrated exceptional accuracy, sensitivity, and specificity for lung cancer versus control, achieving an impressive area under the curve (AUC) of 0.99 in the ROC analysis as shown in Figure 7.This robust performance carried over to the testing set, where all measures reached 100%, indicating a highly reliable diagnostic capability.Likewise, when evaluating gastric cancer versus control, the classification model exhibited remarkable performance in both the training and testing sets, with all performance measures reaching elevated levels.For the more challenging classification model of lung cancer versus gastric cancer, the achieved performance remained notably high, slightly lower than the previous models.Nevertheless, these results were still deemed clinically acceptable.
Additional evaluations were conducted to determine the sensor system's capability to differentiate early-stage (I, II) from late-stage (III, IV) cancers.In the case of lung cancer, the accuracy for the training set was found to be 59%, while for the testing set, it improved to 81%.However, these accuracies were relatively low, possibly influenced by the limited numbers in each subgroup, which might have affected the performance of the classifier.Despite the challenges, the subsequent classification for lung cancer achieved higher accuracy in the validation tests, reaching 71%.While there is room for improvement, these findings demonstrate the feasibility of utilizing the nanosensors for distinguishing early-stage from late-stage cancers.
Chemiresistor sensor arrays with molecularly linked gold nanoparticle thin film assemblies were also tested for detection of mixed VOCs and breath samples from lung cancer patients. [87]he study involved analyzing the response of the sensor arrays to various concentrations of VOCs and breath samples from healthy individuals and lung cancer patients.The sensor arrays used in this study were created by assembling [17,57,88] or printing nanoparticle thin films onto chemiresistor devices.Two methods were used to produce the nanoparticle thin films: self-assembly via exchange-crosslinking-precipitation or print using pre-linked nanoparticle inks. [63,89]Various VOCs and their mixtures under relatively high vapor concentrations were tested for establishing the general sensor response characteristics.The VOCs consist of acetone, hexane, iso-propanol, water, and their combinations which are relevant to the VOCs found in human and lung cancer breaths.The sensor array used includes two sensing films: printed MUA-Au 2 nm and self-assembled MPA-Au 2 nm .The results showed that the MUA-Au 2 nm film demonstrated a more pronounced response to the VOCs compared to MPA-Au 2 nm .Additionally, for the MUA-Au 2 nm , there is a higher response to  Abbreviations: ROC, receiver operating characteristic analysis; AUC, area under the curve.Adapted with permission. [69]Copyright 2020, Wiley.
acetone than hexane, while the opposite trend is observed for MPA-Au 2 nm , which can be accounted for the structure difference between the two sensing films.The sensor testing results for acetone, hexane, iso-propanol, water, and their mixtures reveal distinct separations between each of the groups.Additionally, distinct separations were evident for acetone, iso-propanol, and their mixtures with various mixing ratios.To assess the array's potential for identifying breath samples from lung cancer patients and healthy individuals, a limited number of breath samples were collected from individuals diagnosed with Stage 4 non-small-cell lung cancer and healthy controls.The six patients included in the study all had positive results for adenocarcinoma.The same number of healthy individuals participated.Same breath sample collection protocols were followed by both groups.An optimized array comprising three sensors (sa-MUA-Au 7 nm , sa-BDT-Au 2 nm , and sa-NDT-Au 2 nm ) was used to obtain response data from breath samples of both healthy individuals and lung cancer patients, along with the control experiments, were then subjected to PCA analysis.The PCA plots demonstrated clear separation of the sensor response data between the breath samples of healthy individuals and those of lung cancer patients.
Clearly stated a good differentiation between the lung cancer patients and healthy subjects with 100.0%sensitivity, 83.3% of specificity, and 91.7% of accuracy.The results of this work have implications for the development of highly sensitive and selective chemiresistive sensors for a range of applications, including environmental monitoring, medical diagnostics, and food safety.The ability to detect target molecules at ppb level concentrations could enable early detection of diseases or contaminants, leading to improved outcomes and reduced risk to public health.

Flexible Sensors Based on Nanoparticle-Structured Sensing Interfaces
Nanoparticle-based, multi-parameter sensing platforms [90] are designed to identify, categorize, and segregate common stimuli, like VOCs.Such sensors hold promises for applications in wearable devices, humanoid robotics, and health monitoring systems.The flexible sensors were fabricated by drop-casting solutions of gold monolayer-capped nanoparticles (Au-MCNPs) onto Kapton substrates, with Cu/Ag interdigitated electrodes, or onto polyethylene terephthalate (PET) substrates with Ag electrodes.A series of chemiresistive films were fabricated using Au-NPs coated with various alkanethiols, specifically 1-butanethiol (C4), 1-hexanethiol (C6), 1-decanethiol (C10), and 1-octadecanethiol (C18).These films were meticulously prepared to investigate the impact of ligand chain length on the performance of flexible Au-MCNP chemiresistors.Within a specially designed chamber, the sensors were strategically positioned to allow bending when exposed to VOCs.Subsequently, these sensors were exposed to varying concentrations of different VOCs, exhibiting responses to mesitylene.In general, as the concentration of VOCs increased, an overall rise in sensor response occurred.For Au-MCNPs coated with relatively shorter chain lengths, such as C4 and C6, the responses were positively correlated with concentration.Conversely, Au-MCNPs coated with longer chain lengths, like C10 and C18, exhibited a negative response trend.The sensor response (ΔR/R 0 curves) for mesitylene is examined.The sensitivity for each VOC was determined by calculating the slope of the linear fit between the relative response (ΔR/R 0 ) and the P/P 0 value (where P represents the partial pressure of the VOC and P 0 signifies the vapor pressure of that specific VOC).Notably, Au-NPs capped with 1-butanethiol demonstrated the highest sensitivity to mesitylene.The films composed of Au-NPs capped with C10 and C18 exhibited negative responses when exposed to cyclohexanol.
The concept of a single flexible sensor platform with the ability to function as multiple sensors is evaluated by exploiting additional sensing characteristics that emerge during deformation of the sensing layer.Specifically, seven sensors made of Au-NPs capped with hexanethiol were fabricated with an average baseline resistance of 45.3 MΩ ± 7.4 MΩ without deformation.Six of these sensors were tested with different bending degree with radii of 0.21, 0.26, and 0.28 mm − ¹.Half of the bent sensors were configured so that the sensing film underwent stretching during deformation, as illustrated in Figure 8A, while the other half (D) Specific VOCs separation.Adapted with permission. [90]Copyright 2017, Wiley.
was bent to subject the film to compression deformation.These flexible sensors were then fixed in a static bent position and introduced into a chamber, where they were exposed to seven different VOCs at increasing concentrations.PCA analysis was conducted based on the sensor responses.Remarkably, upon organizing all the exposure data into bending groups, an orthogonal classification was discovered as illustrated in Figure 8B.This classification presents captivating possibilities, suggesting that a flexible sensing platform possesses the capability to not only detect when it undergoes deformation but also discern the precise direction of that deformation, even within a multifaceted environment.Additionally, a conspicuous grouping pattern emerges when contrasting polar and non-polar analytes, as seen in Figure 8C, with additional grouping opportunities available for specific analytes, as demonstrated in Figure 8D.The utilization of flexible sensors based on Au-MCNPs, each encapsulated with a single type of ligand, enables the attainment of multiparametric sensing capabilities for various VOCs and the assessment of bending states such as stretching and compressing.

Flexible Sensors for Detecting Both VOCs and Strains
Another focal area is the development of flexible sensing platforms based on nanoparticle-structured sensing interfaces for detecting both VOCs and device strains.A flexible chemiresistor is fabricated by utilizing a microelectrode-patterned polyethylene terephthalate (PET) as a substrate and coating it with a thin film assembly of gold nanoparticles. [60]The devices are examined under varying device strains and in the presence of VOCs.The sensor's response dependencies on the chemical properties and vapor composition are delineated.Simultaneously, from the viewpoint of device flexibility, how the dependence aligns with the device's strain characteristics is examined.Consequently, this study comprehensively investigates both the sensor's reactivity concerning vapor composition and its response to device bending.The response characteristics across various device strains are characterized by differing radii of curvature, as well as distinct vapor exposure environments.The choice of these specific vapors was predicated upon disparities encompassing a blend of molecular polarity, hydrophobicity/hydrophilicity, and dielectric constant properties.Figure 9 illustrates a representative set of sensor response profiles to hexane, ethanol, and acetone when tested on a device featuring a thin NDT-Au 2nm film under varying device strains.Analyzing the hierarchy of differences in response sensitivities to these distinct vapors, which follows the sequence ethanol > hexane > acetone (Figure 9A-C), suggests that alterations in the dielectric characteristics of the thin film due to vapor sorption likely exerted a predominant influence on the corresponding changes in the electrical properties of the sensing film.These data reveal that the variations in resistance relative to vapor absorption are more influential than the response to device strain at the observed vapor concentrations.This finding underscores that within the range of tested vapor concentrations, the sensor's linear response to vapor concentration remains largely unaffected by the device's strain. 2 cm (ts), (c) 0.9 cm (ts), (d) 0.7 cm (ts), (e) flat, (f) 1.2 cm (cs), (g) 0.9 cm (cs), and (h) 0.7 cm (cs).Note: On: vapor purging, Off: N 2 purging.Vapor concentration (ppm (M), from left to right): (1) ethanol: 322, 644, 966 and 1288; (2) hexane: 826, 1652, 2478, 3304; and (3) acetone: 1259, 2518, 3778, 5037.D) Illustrations of the different device bending directions and degrees.Adapted with permission. [60]Copyright 2014, RSC.
Regarding the sensor's reactivity to varying degrees or directions of strain, it is notable that the discrepancies in response sensitivity are predominantly modest, particularly within the lower vapor concentration range.This discovery holds significance, suggesting that the sensor's responsiveness remains relatively unperturbed by the bending parameters when operating under conditions of low vapor concentrations.Nevertheless, upon closer examination of the slopes, subtle distinctions emerge among the various vapors and bending orientations.Specifically, when comparing the selectivity differences, we observe approximately 2.6%, 4.2%, and 4.9% variations in the case of tensile strain for hexane, ethanol, and acetone vapors, respectively.In contrast, for compressive strain scenarios, we note differences of −3.2%, −0.9%, and 5.2% in the corresponding directions for the same vapors, i.e., hexane, ethanol, and acetone.Through the comparative analysis of the response sensitivities exhibited by the flexible chemiresistor under varying strain conditions in the presence of hexane, ethanol, and acetone vapors, three key findings have emerged.To begin, it is evident that alterations in the dielectric characteristics of the thin film due to vapor sorption exert a dominant influence on the electrical properties of the nanoparticle thin film.Secondly, minimal disparities in response sensitivity to different strain directions or degrees, particularly within the lower vapor concentration range, are observed, especially when dealing with vapors with low dielectric constants.A substantial tunability in response sensitivity is demonstrated.

Detection of VOCs with Nano-filamented Textile Sensor Platform
Another example includes an electrospun polymeric fibrous substrate embedded with nano-filaments defined by sizetunable gold nanoparticles and structurally sensitive dendrons as crosslinkers. [65]The size-tunable gold nanoparticles and dendrons assemblies were prepared through an exchangingcrosslinking-precipitation method.The electrospun polymeric fibrous substrate used in this work was the PAN/PET fibrous membrane substrates and was prepared via a method as previously described. [91]The fibrous chemiresistive sensors used in this work contain three major components.The fibrous substrate was considered as a chemiresistive component and was integrated with carbon-printed interdigitated microelectrodes (IMEs) which served as a conductive filament.Subsequently, the dendron-Au nanoparticles with different sizes (diameters of 2,6,13, and 30 nm) were assembled as the sensing interface.
Methanol was chosen to assess the sensor response of a dendron-Au NPs-based chemiresistive sensor on a fibrous paper substrate.As the nanoparticle sizes increase the absolute values of sensor responses demonstrated an increment as well.Furthermore, the response profile exhibited a shift from positive to negative when the nanoparticle sizes reached 13 and 30 nm.Other alcohols displayed similar response characteristics.Upon comparing the sensitivity data of n-alcohols and iso-alcohols, the response profile was significantly influenced by the particle sizes, where smaller nano filaments exhibit lower sensitivity and positive response while larger ones display an opposite trend with higher sensitivity and negative response as shown in Figure 10A.Additionally, the study unveiled noticeable variations in response sensitivities between n-alcohols and iso-alcohols.To eliminate the influences of different synthesis methods on the sensing properties, data obtained from organic-synthesized Au NPs (6 nm) was compared with the data from aqueous-synthesized Au NPs (6 nm).In both cases, positive response profiles and a similar chain length effect were observed.
This type of nano-filamented sensor displays distinct structure and response characteristics compared to previous chemiresistive sensors using thin film assembly of alkanethiolate-capped  B,C) 2D and 3D PCA plots for the array of six selected sensors with six different nano filaments (DT-AuNP (6 nm) and DD-AuNP (6 nm), and fibrous DD-AuNP (2, 6, 13, and 30 nm) responses to n-alcohols and their isomers.Adapted with permission. [65]Copyright 2022, ACS Publication.
Au NPs.Unlike those sensors that predominantly generate positive response profiles due to VOC adsorption and resistance increase caused by interparticle distance expansion upon vapor sorption. [57]Examples include Au NPs assembly with different crosslinkers, such as aromatic organo-thiol derivatives (HS-C 6 H 4 -X), when the functional groups (X) are ─OH and ─CH 3 , the sensors exhibit positive responses to methanol and negative responses when X is ─COOH or ─NH 2 . [92]When hydrogen-bonding mediators are involved in the assembly of Au NPs, the sensors display negative responses to water and methanol but positive responses to ethanol, propanol, and other similar compounds. [93]The surface chemistry of gold, involving Au-thiolate interactions, allows for the self-assembly of thiolate monolayers with various functional groups.This capability enables the tuning of sensing interfaces for different applications. [94,95]This array served as a compelling proof-ofconcept, illustrating its remarkable capacity for detecting alcohol molecules with various chain lengths and isomers, exhibiting high structural sensitivity.The sensor array consists of six sensors derived from nano filaments with variations in their substrates (flat and fibrous), nanoparticle sizes (diameter of 2, 6, 13, and 30 nm) and linker molecules of varied chemical structures, which are sensors derived from 2D rigid decanethiol-Au 6nm and dendron-Au 6nm , and 3D fibrous dendron-AuNP (2, 6, 13, and 30 nm).
PCA was applied to analyze the array's response sensitivities to alcohols with diverse chain lengths, the findings revealed a remarkable level of selectivity.This selectivity was not only evident for alcohols with varying chain lengths but also for distinguishing between different isomers, specifically n-alcohols and isoalcohols.The statistical method Random Forest classification was employed to provide feature importance for the sensors.Among all the sensors, 2D rigid dendron-Au 6nm , and 3D fibrous dendron-Au 13nm , dendron-Au 30nm , NDT-Au 30nm have the highest importance feature and were selected for a sensor array.This sensor array displayed a similar level of alcohol separation as the previous 6-sensor array as illustrated in Figure 10B,C.Based on the PCA results, this selected 4-sensor array had the capacity to differentiate alcohols with different chain lengths evidenced by the gradual shifts of PCA coordinates.Moreover, each cluster is clearly separated in the PCA plot, it is evident that the array is able to distinguish between normal alcohols and their respective isomers.
The multiplexing capability exhibited by the nano-filamented sensors is crucial for detecting various isomeric alcohol VOCs present in lung cancer breath or other biomarkers, compounds like 2-ethyl-1-hexanol, 2-ethyl-4-methyl-1-pentanol, and 2-propyl-1-pentanol, which are often challenging to differentiate under high humidity conditions. [6]Conventional methods for detecting blood alcohol content levels usually involve invasive and time-consuming procedures such as blood sampling or urine .Adapted with permission. [54]Copyright 2023, ACS Publication.
monitoring.However, utilizing saliva, breath, and sweat samples offers a more convenient and noninvasive approach for real-time alcohol detection.Nevertheless, most of the sensors detected human fluids suffer from a significant false-positive rate, primarily attributed to interference caused by high moisture content and other compounds present in the samples.Furthermore, despite the availability of noninvasive sensors, including commercial options, none have been able to exhibit the capability of effectively sensing alcohols with various chain lengths and isomers.The nanofilamented polymeric network arrays offer enhanced adjustability in terms of sensitivity and selectivity, making them promising candidates for various applications.These arrays hold potential in breath sensing for alcohol VOC biomarkers related to lung cancer and in sweat sensing for detecting other biomarkers, even under challenging high humidity or moisture conditions.

Detection of Human breaths with Simulated Lung Cancer Breath VOCs
Built upon the excellent performance of the nanostructured sensor arrays for the breath VOC sensing, a compact and wireless breath sensor system was recently demonstrated. [54]This integrated system consists of sensor electronics, breath sampling components, data processing, and sensor arrays based on nanoparticle-structured chemiresistive sensing interfaces (Figure 11A) which is used to detect VOCs that are relevant to lung cancer biomarkers in human breath samples.The eightsensor array's responses to human breath samples, both with and without lung cancer-specific VOCs, are subjected to PCA analysis.The first two principal components (PC1, PC2) were extracted from this analysis, which serves as the classification features for further evaluation and identification.In Figure 11A, the PCA plot depicts the sensor response data from human breath samples.The clusters on the plot are noticeably separated, with one group representing healthy breath samples and the other representing lung cancer breath samples.While there is a slight overlap between the two groups, it is primarily due to the limited sample size of the study.
To evaluate the sensitivity and selectivity of the collected data, a comprehensive data analysis is conducted using multiple neural network modules with the Random Forest (RF) algorithm for pattern recognition.By examining the PCA results, various performance metrics such as classification rate, sensitivity, and selectivity are assessed.The Random Forest classifier demonstrates an accuracy of 93.99% when tested against the data.Furthermore, it successfully identifies 86.96% of the lung cancer breath samples and with a 100% accuracy in classifying healthy breath samples, signifying its effectiveness in distinguishing between healthy and lung cancer samples.The sensor responses to human breath samples, both with spiked lung cancer-specific VOCs ("lung cancer") and without ("healthy"), were investigated for their stability.Data was collected from a series of continuous 6 d human breath tests, and the sensor array response data for each day was subjected to PCA analysis as shown in Figure 11B-D.The results clearly showed a noticeable distinction between the data from healthy breath samples and those spiked with lung cancer VOCs.Throughout the six-day continuous testing, there is minimal overlap between the two categories, with only a slight overlap observed on the last day.This outcome serves as strong evidence of the high stability of the sensor responses over time.To further evaluate the stability of the sensor array, the average and standard deviations of the sensor responses toward the human breath samples were analyzed.The results obtained from repeated tests demonstrated a minimal standard deviation at the magnitude of 10 −4 , a relative standard deviation (RSD) of 0.627% Figure 12.Illustration of the system-level approach to sensor platform by integration of the various components.Adapted with permission. [55]Copyright 2023, IEEE for sensor responses to human breath samples and 0.542% for human breaths spiked with lung cancer-specific VOCs.This consistency reinforces the sensor array's reliability and robustness over time.

Integration of Sensor Array System and Application of AI for Data Analysis
The establishment of an effective approach for the integration of nanostructured chemiresistive sensing arrays with data analysis methods is of paramount significance, particularly in the context of creating a portable and wireless sensor platform. [55]The synergy between the sensor array and artificial intelligence (AI)powered data analysis facilitates the proficient gathering of data and the construction of a comprehensive database.The sensor array system initially acquires data from human breath samples, which are subsequently transmitted and uploaded wirelessly to a cloud-based database.After a predefined interval, these data can be retrieved for the purposes of data training and in-depth analysis.A wireless and portable platform is shown in Figure 12.This integrated system comprises sensor hardware and electronics, facilitating data collection from the chemiresistive sensor array responses to VOCs and human breath samples.Furthermore, the platform seamlessly transfers and stores the acquired data in a cloud database for efficient management and analysis.In order to assess the effectiveness of the integrated system for remote monitoring of human breath samples, two separate sensor array stations were established in different physical locations.The study involved five volunteers, and the sensor testing phase was extended for over 22 d.Based on the preliminary data collected, the system exhibited remarkable stability and continuity throughout the study, illustrating its capacity to capture distinct patterns from the same individual on different days.Moreover, the successful wireless transfer and storage of data in a cloud database further validated the system's efficiency and feasibility in realworld applications.
Subsequently, this data was subjected to further analysis for response patterns, providing valuable sensing information for evaluating multiple analytes.The extracted information is then processed for subsequent machine learning analysis, allowing for more in-depth insights and potential applications.In this study, a particular example of remote screening involving human breath samples with and without simulated lung cancer-specific VOCs is examined.The dataset used in this example consists of 150 data points, with 75 samples representing healthy breaths and another 75 simulated breaths from lung cancer patients.These data points were divided into a training set of 70% and a testing set of 30% for conducting the analysis.During the sample testing phase, data collection occurred over three consecutive days.50% of the data was collected five hours after sample collection, while the other half was collected after 24 h of sample collection.This approach was meant to assess the performance of the sensor array system concerning variations in sample shelf time, providing valuable insights into the system's capabilities under different conditions.PCA analysis is conducted based on the collected data.The outcome reveals two distinct clusters, with only a small overlapping area between them.This clear separation indicates the sensor array's effectiveness in distinguishing between human breath samples with and without spiked lung cancer-specific VOCs, highlighting its potential for accurate VOC detection.
The performance evaluation of various analytic methods (Figure 13A-E) was conducted as well, focusing on sensitivity and selectivity.Figure 13F-J presents a representative set of confusion matrices, displaying the classification results obtained from different Machine Learning techniques.In these matrices, the diagonal elements represent the total correct predictions per class.The color coding is used to highlight the model's performance, with darker shades indicating higher accuracy.Additionally, the Figure 13.A-E) Schematic of statistical methods used for data analysis (A-E: KNN, ANN, SVM, XGboost, and RF classifier respectively).(F-J) Confusion matrix result of (F) KNN Classifier, (G) SVM Classifier, H) RF Classifier, I) XGBoost Classifier, and J) ANN Classifier.Samples: human breath (HB) and HB with lung cancer-specific VOCs.Adapted with permission. [55]Copyright 2023, IEEE.
differentiation between darker and lighter colors on the diagonal elements illustrates the model's effectiveness in terms of sensitivity and selectivity when identifying healthy and lung cancer-VOC-spiked human breaths.To evaluate the sensitivity and selectivity of the sensor array data, five statistical methods were employed: XGBoost, Random Forest (RF), artificial neural network (ANN), k-nearest neighbor (KNN), and support vector machine (SVM).Among these methods, the XGBoost classifier outperformed the others, achieving an overall correct classification rate of 82%.Specifically, it successfully identified 84% of the breath samples containing lung cancer-specific VOCs and a selectivity of 81%.This portable and wireless sensor platform with an integrated a nanostructured chemiresistive sensing array and sensor electronics shows indeed promises for potential point-ofcare and remote monitoring applications, specifically for lung cancer screening through human breath analysis.

Summary and Future Perspectives
In summary, the chemiresistive sensors with nanostructured sensing interfaces possess the tunability for the detection of VOCs from human breath.Significant progress has been made in various aspects of this research front, ranging from the development of sensing materials, with a focus on nanomaterials and hybrid structures, to the emergence of advanced sensing systems and their diverse applications.These studies highlight the potential for significant advancements in the field of sensing technology and its wide-ranging applications.As we move forward, the focus for the next generation of chemiresistive sensors should be on coupling the enhanced sensitivity and selectivity with artificial intelligence.This can be achieved through the adoption of advanced materials and structures or by effectively extracting target signals using advanced machine learning algorithms, paving the way for more accurate and reliable VOC sensing solutions that cater to a wide range of critical applications.Indeed, the nanostructure-based hybrid sensor array was recently used for the detection of COVID-19 in exhaled breath. [68]This work highlights a noninvasive approach to identifying and monitoring individuals who are at risk or already infected.The breath analysis device is designed with a nanomaterial-based hybrid sensor array with multiplexed detection capabilities.The array incorporates different gold nanoparticles with organic ligands. [69]A pioneering clinical study was undertaken, involving 140 participants.Statistical analysis further aids data comparisons, demonstrating remarkable sensitivities and good selectivity.This highlights its potential as a reliable and comprehensive diagnostic tool for virus detection.The results obtained in this study show a comparable level of accuracy when compared to existing published data. [96,97]The mass-screening protocols and technologies presented in this work offer significant advantages, as they substantially reduce testing time and facilitate immediate data sharing, making them valuable tools for efficient virus screening and diagnosis.Furthermore, chemiresistive sensors hold great promise in the field of wearable applications.The utilization of nanoparticlenanofibrous membranes as tunable interfacial scaffolds for flexible sweat sensors [65,98] entails assembling gold nanoparticles within a multilayered fibrous membrane structure with flexibility and adaptability, allowing for wearable and reliable applications.The assembly or printing of molecularly-linked gold nanoparticle thin films onto flexible chemiresistor devices also offers the opportunities for the creation of a flexible sensor device capable of exhibiting high and anisotropic gauge factors for wearable electronics and skin sensors. [89]Similar approaches have been demonstrated by the deposition of cross-linked gold nanoparticles onto plasma-treated PE substrates using a layer-by-layer selfassembly technique, [99] leading to multifunctional strain gauges and chemiresistors.While nanoparticle-based flexible sensors hold immense promise across a vast array of applications, [100,101] considerable efforts are being made to further increase the sensitivity, selectivity, and durability.Nanoparticle-based flexible sensors are poised to emerge as one of the foremost and transformative applications of nanotechnology, which is expected to aid the design of advanced chemiresistive sensors to meet the demands of various real-world wearable applications, including rapid breath screening of disease and environmental monitoring of toxic gases.Moreover, significant advancement is anticipated in point-of-care and point-of-need screening and diagnostic applications.

Figure 1 .
Figure 1.Illustration of the integration of the detection electronics and data processing with chemiresistor sensor arrays derived from nanoparticle-structured sensing interfaces for detection of human breath VOC biomarkers for screening and diagnostics.

Figure 2 .
Figure2.PCA plots based on simulated chemiresistor response data for two different panels of VOCs (A, B) found in human breaths (healthy and lung-cancer patient breaths) on a sensor array derived from nanoparticle assemblies with different parameters of the sensing nanostructures (VOCs concentration in with 1-40 ppb and 400 ppb-7 ppm (V) ranges).Adapted with permission.[54]Copyright 2023, ACS Publication.

Figure 4 .
Figure 4. Illustration of two approaches to sensor fabrications: A) Fabrication of strain sensors by nanoink-printing and pulsed-laser sintering of SANs from AuCu NPs.B) Pulsed-laser sintering on PET substrate.C) SMI-SMA processes for fabrication of flexible fibrous strain sensors.D) Illustration of printing microelectrodes by SMI (top) and printing sensing film by SMA (bottom).Adapted with permission.[62]Copyright 2021, Elsevier.

Figure 5 .
Figure 5. A) Illustration of the 3D fibrous textile sensor interface.B) Illustration of the functionalized Au NP coated fibers as sensitive filament and carbon-coated fibers as conductive filament.C) Illustration of tunability of the sensitive filaments in terms of Au NP and fiber sizes.D) Illustration of tunable sensitive filament in terms of interparticle ligands and linkers.(E) Illustration of a chemiresistor sensor array.(DD stands for dendron; NDT for 1,9-nonanedithiol, and DT for decanethiolate.).Adapted with permission.[65]Copyright 2022, ACS Publication.

Figure 7 .
Figure 7. A) Data classification of the clinical trial based on leave-one-out cross-validation through discriminant factor analysis of the sensor array results.Box plots on the first canonical score of the training set (blue square) and testing set (red star) for lung cancer versus Control.B) ROC analysis of the training set including the AUC for lung cancer versus control.The horizontal dashed line in the box plots represents the cut-off value of the model.Abbreviations: ROC, receiver operating characteristic analysis; AUC, area under the curve.Adapted with permission.[69]Copyright 2020, Wiley.

Figure 8 .
Figure 8. A) Illustration of Au-MCNPs deposited on a flexible substrate while being bent to banding radii of 0.21, 0.26, and 0.28 mm −1 .PCA plot based on 2 features extracted from the responses of Au-MCNPs based sensors to 1-hexanetiol as encapsulating ligand in response to 7 VOCs.B) Orthogonal pattern displayed for stretching (bend+) compressing (bend-) and flat (Bend0) conditions of the flexible sensors.C) Polar versus non-polar separation.(D)Specific VOCs separation.Adapted with permission.[90]Copyright 2017, Wiley.

Figure 10 .
Figure 10.Response sensitivities and PCA plots of the sensor responses for the sensitive filaments with different Au NP sizes.A) Response sensitivities versus number of carbons in n-alcohols (empty bars) and iso-alcohols (filled bars).B,C) 2D and 3D PCA plots for the array of six selected sensors with six different nano filaments (DT-AuNP (6 nm) and DD-AuNP (6 nm), and fibrous DD-AuNP(2, 6, 13, and 30 nm) responses to n-alcohols and their isomers.Adapted with permission.[65]Copyright 2022, ACS Publication.

Figure 11 .
Figure 11.A) Illustration of the sensor testing system.B-D) PCA plots of sensor array responses to human breath samples with and without spiked lung cancer-specific VOCs over a period of consecutive 6 d (B: day1, C: day 3, and D: day 5).Adapted with permission.[54]Copyright 2023, ACS Publication.

Table 1 .
Examples of chemiresistive sensors for VOC detection.