Metal Oxide Semiconductor Gas Sensors for Lung Cancer Diagnosis

Abstract

Lung cancer is the most prevalent severe illness in both sexes and all ages and the leading cause of cancer-related deaths globally. Late-stage diagnosis is the primary cause of its high mortality rate. Therefore, the management of lung cancer needs early-stage screening. Breath analysis is a non-invasive, low-cost, and user-friendly approach to diagnosing lung cancer. Among the various types of breath sensors, MOS gas sensors are preferred due to their high gas responses, fast response times, robustness, and lower price. This review focuses on the critical role of MOS gas sensors in detecting VOCs in lung cancer patients’ exhaled breath. It introduces the basic working mechanism of MOS gas-sensitive materials, summarizes some high-performance MOS materials suitable for detecting potential lung cancer biomarkers and provides performance enhancement strategies. The review also briefly introduces the sensor array and its pattern recognition algorithm. Finally, we discuss the challenges in developing MOS gas sensors for lung cancer screening and present the prospect of using the e-nose for large-scale early lung cancer screening.

1. Introduction

Cancer incidence and mortality rates have risen rapidly worldwide, making it a significant cause of death for the global population [1,2]. The latest global cancer statistics report, “Global Cancer Statistics 2020”, evaluated data on 36 major cancers from 185 countries or regions worldwide, revealing nearly 19.3 million new cancer cases in 2020, with lung cancer accounting for over 2.2 million cases and approximately 1.8 million deaths [2]. Of all cancer types, lung cancer has the highest mortality rate of 18%, posing a severe threat to human health and life (Figure 1). The five-year survival rate for lung cancer patients is only approximately 15% [3], with the prognosis varying significantly based on the clinical stages [4,5]. The TNM (tumor-node-metastasis) staging system (version 8) suggests that early-stage I lung cancer patients can achieve a five-year survival rate of over 70% after surgical treatment, while late-stage IV lung cancer patients remain below 10%, even with adequate treatment [4]. Early diagnosis and treatment are critical to improving the prognosis of lung cancer patients [6]. However, early lung cancer rarely exhibits clinical symptoms, and nonspecific symptoms are the leading cause of obtaining a lung cancer diagnosis [7,8,9]. As a result, lung cancer is often diagnosed in late-stage, and the cancer tissue has already metastasized from the primary site, indirectly leading to the high mortality rate of lung cancer [10]. Therefore, active screening is crucial in diagnosing early-stage lung cancer, significantly reducing treatment pain, improving survival rates, and avoiding the high costs of late-stage treatment.

Currently, clinical lung cancer diagnosis mainly relies on sputum cytology examination [11,12], histopathological examination [13,14,15], and imaging examination [16,17,18,19,20,21]. Sputum, secreted from the lungs, bronchi, and trachea, can carry pathological information when these areas are diseased [11,12]. Despite its simplicity and low equipment requirements, a cytology examination of exfoliated cells in the sputum is limited by its identification of cell morphology and high false positive rate, which depends on the quality of the sputum specimens and physician experience [12]. Histopathological examination, obtained through percutaneous puncture lung sampling, provides a reliable lung cancer diagnosis with high diagnostic accuracy and staging confirmation [15]. However, histopathological examination is typically conducted after a lesion has formed due to its high invasiveness and reliance on accurate localization, rendering it unsuitable for early screening in high-risk populations. Bronchoscopy can visualize the lesion location and provide a qualitative diagnosis by inserting a fiberoptic probe into the bronchus, but it is limited by the probe’s reach and observation range and is unsuitable for early lung cancer screening due to the difficulty of seeing early lesions under the mucosa [13,14]. Imaging examination using CXR (Chest X-ray), CT (computed tomography), and PET (positron emission computed tomography) has become a commonly used early lung cancer screening method to quickly and non-invasively examine specific areas to show the position, size, and shape of the tumors. However, each modality has limitations, such as CXR’s limited resolution for detecting only larger lumps (greater than 10 mm) [16] and LDCT’s (low-dose CT) higher false positive rate [17,18,19]. PET has high sensitivity and specificity for early-stage lung cancer diagnosis but is costly, making it challenging to be popularized in large-scale routine examinations [20,21]. Additionally, radiation exposure is a practical concern in all imaging examinations.

Breathing is the physiological gas exchange process between the human body and the external environment. The human metabolism generates various volatile organic compounds (VOCs) that can pass through the bloodstream to the alveoli and exhale via respiration. As such, the VOC composition in exhaled breath can indicate human metabolism and health status [22]. Consequently, by detecting and analyzing the components of exhaled breath to identify lung cancer biomarkers, exhaled breath analysis is a promising tool for early lung cancer screening [23,24,25,26]. Compared to conventional lung cancer diagnostic methods, exhaled breath analysis is non-invasive, convenient, and provides rapid and accurate results [27,28].

As early as the 1970s, Pauling et al. employed gas chromatography (GC) to identify approximately 250 VOCs in the exhaled breath of healthy individuals [29]. Since then, advances in detection methods have led to identifying a more significant number of VOCs. Using gas chromatography–mass spectrometry (GC–MS), Phillips et al. detected 3481 VOCs in exhaled breath samples from 50 healthy subjects, of which only 27 VOCs were found in all subjects [30]. The study revealed that acetone, isoprene, ethanol, methanol, and 2-propanol were the major VOCs in exhaled breath, with 2-propanol being considered exogenous due to its similar concentration in exhaled breath and ambient air [31,32]. Inorganic gases, such as N₂, O₂, CO₂, and H₂O, are the main components in exhaled breath, potentially interfering with identifying VOCs [32]. These findings indicate that the exhaled breath composition is complex, and identifying VOCs as lung cancer biomarkers remains challenging.

In 1985, Gordon et al. first used GC–MS to analyze the exhaled breath samples of 17 healthy individuals and 12 lung cancer patients and found 22 VOCs that differed between the two groups. They first proposed using exhaled breath for lung cancer diagnosis [26]. Subsequently, Phillips’ team discovered that the exhaled breath of lung cancer patients contained significantly more alkanes and emphasized the potential of VOCs in exhaled breath for lung cancer diagnosis [33,34,35]. Additionally, they developed nonlinear multivariate analysis methods that improved the ability to diagnose lung cancer by utilizing different combinations of VOCs in exhaled breath [36,37]. O-toluidine [38], toluene [39], and 1-propanol [40] were commonly regarded as VOC biomarkers for lung cancer, as their concentrations were found to be significantly higher in the exhaled breath of lung cancer patients than in that of healthy individuals. Koureas et al. found a strong correlation between the presence of ethylbenzene, toluene, styrene, 2-propanol, and 1-propanol in exhaled breath and the ability to distinguish between cancer patients and healthy individuals [41]. Schallschmidt et al. quantified the concentration of 24 potential VOC biomarkers for lung cancer [42]. They found that the concentration of aromatic compounds in the exhaled breath of smokers increased. In contrast, the oxygen-containing VOC concentration in lung cancer patients’ exhaled breath increased significantly, such as aldehydes, 2-butanone, and 1-butanol. The production of these VOCs may be related to the oxidative stress behavior of lung cancer cells caused by excessive free radicals and reactive oxygen species (ROS) in the cells of lung cancer patients [43,44,45].

Figure 2 summarizes the hypothetical basis for generating different types of VOC biomarkers in the exhaled breath of lung cancer patients [46]. During oxidative stress, ROS can cause oxidative DNA damage, polyunsaturated fatty acids (PUFAs), proteins, and other substances in the body, while also generating volatile hydrocarbons and aldehydes, which excrete through respiration [46,47,48]. Exposure to carcinogens (such as cigarette smoke, benzene, and nitrosamines) can increase the risk of cell carcinogenesis by causing ROS accumulation. It induces changes in cytochrome P450 enzyme activity patterns, generating VOC biomarkers for lung cancer [46]. The Warburg effect, resulting from the prompt proliferation of lung cancer cells in a hypoxic condition, is another possible source of VOC biomarkers [49]. At this time, glycolysis activity is higher than oxidative phosphorylation, increasing the metabolites of glycolysis, such as acetaldehyde, ethanol, and acetone, in exhaled breath [50,51].

However, there is no consensus on lung cancer biomarkers in exhaled breath due to the variations in breath sampling methods, gas detection technologies, and patient factors (including lung cancer type, body weight, and dietary habits) [52,53,54]. Jia et al. compared the VOC types in exhaled breath from lung cancer patients and the in vitro lung cancer cell cultures and discussed the methodological issues that may cause inconsistencies between studies. They suggested that potential VOC biomarkers for lung cancer may include 1-propanol, isoprene, acetone, pentane, hexanal, benzene, toluene, and ethylbenzene [55]. Schmidt et al. reviewed research on the exhaled breath biomarkers for lung cancer over the past four decades and proposed using frequently studied VOCs as potential biomarkers [56]. Figure 3 shows some of the most common VOCs identified as lung cancer biomarkers, including BTEX (benzene, toluene, ethylbenzene, and xylene), isoprene, hexanal, nonanal, 2-butanone, acetone, pentane, and 1-propanol [56]. Notably, no VOCs have been identified that are exclusively present in exhaled breath from lung cancer patients [42,57].

In order to assess the reliability of the diagnostic methods utilizing exhaled breath, researchers have employed machine learning and selected various combinations of potential lung cancer VOCs to build prediction models [8,41,53,58]. For example, Chen et al. developed a model comprising 20 VOCs that achieved 93.9% accuracy in distinguishing between non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC), and another model composed of 19 VOCs that could differentiate between early and late-stage lung cancer with an accuracy of 82.7% [8]. Nevertheless, the most prevalent method used for exhaled breath diagnosis entails the application of large instruments such as GC–MS. While effective in detecting low-concentration VOCs, they have limitations, such as complexity, high costs, and lack of portability, hindering their widespread use in routine home check-ups. Furthermore, they cannot provide real-time analysis. In a prospective study, Peled et al. used GC–MS and a gas sensor to detect the exhaled breath in 72 patients with lung nodules, and the latter selectively distinguished between benign and malignant nodules via pattern recognition, even successfully differentiating between two different types of lung cancer (adenocarcinoma and squamous cell carcinoma) [54], indicating that gas sensors have the potential to become low-cost, small, portable exhaled breath detection and analysis devices. Advanced gas sensors have been developed for low-concentration gas detection in recent years [54,59,60,61,62,63,64,65], including surface acoustic wave (SAW) sensors, quartz crystal microbalance (QCM) sensors, electrochemical gas sensors, and oxide semiconductor (MOS) gas sensors. In particular, the MOS gas sensors have made them a promising alternative for VOC detection due to their small size, low cost, fast response, and sensitivity in low-concentration (ppt levels) gases [48,66].

Many reviews about applying MOS gas sensors in exhaled breath disease diagnosis exist. These studies have focused on various aspects, such as identifying a single gas (such as acetone) [67,68], developing sensor devices [69,70], utilizing MOS materials with special structures (such as MOF) [71,72], and improving sensitivity and selectivity [73]. However, despite the potential of MOS gas sensors in detecting VOCs in exhaled breath [60,61,62], there is currently no universal marker for lung cancer. Considering the urgency for early lung cancer diagnosis and the significant potential of MOS gas sensors in this area, it is necessary to review the recent progress in this field. This review summarizes the latest research on MOS gas sensors in lung cancer diagnosis, including their working mechanisms, candidate MOS materials, and the current research progress of MOS sensor arrays. We also discuss the challenges in developing MOS gas sensors for early lung cancer screening.

2. MOS Gas Sensor for Lung Cancer Biomarker VOCs

MOS gas sensors are highly regarded for their high sensitivity, fast response, simple fabrication, durability, small size, and easy integration [48,74]. The origin of semiconductor gas-sensitive materials dates back to 1931 when Engelhard et al. discovered that the conductivity of Cu₂O changed with water vapor adsorption. Despite this finding, it received little attention [75]. In 1962, Seiyama et al. first observed the distinct resistance of ZnO thin films in combustible gases and air at high temperatures, establishing the foundation for MOS gas sensors [74]. Since then, MOS gas sensors have evolved rapidly, with the first commercial MOS gas sensor developed in 1968. To satisfy the requirements of MOS gas sensors in complex environments, Persaud et al. proposed a method of mimicking the animal olfactory system (e-nose) in 1982 to improve the selectivity of sensors for VOCs by using an array of MOS materials with different characteristics to detect mixed gases concurrently [76]. In the following decades, the gas-sensitive mechanism of MOS was further explored [77,78,79,80], the gas sensitivity of a single MOS was continuously enhanced [81], and the fabrication methods of sensor arrays and analysis algorithms were constantly being improved [82,83].

2.1. Working Mechanism

The working mechanism of MOS gas sensors depends on the sensing material’s conductivity change when exposed to different gas environments, enabling target gas detection. Several theories, including the chemisorbed oxygen model [78], the grain boundary barrier model [79], the bulk resistance model [84], and the space-charge layer model (electron depletion layer (EDL) and hole-accumulation layer (HAL)) [77,80], can explain the MOS conductivity changes. The key to these theories is the interaction between the gas and material surfaces [85]. When the affinity energy of the gas molecules exceeds the work function of the MOS surface, electrons transfer from the MOS surface to the gas molecules, resulting in gas anions forming and being adsorbed onto the MOS surface [86]. In particular, the chemisorbed oxygen species (O₂⁻, O⁻, O²⁻) play a crucial role in MOS surface conductivity [78]—closely related to the working temperature and MOS type—determining the MOS’s gas-sensing properties [87,88,89]. The formation process of chemisorbed oxygen species on the surface of SnO₂ can be summarized by the following formulas [89] (Formulas (1)–(4)). We should note that MOS gas sensors have a broad response mode, leading to low selectivity, which can be improved by compounding them with other types of materials [90].

$$\mathrm{I}\mathrm{n}\mathrm{a}\mathrm{i}\mathrm{r}:{\mathrm{O}}_2\left(\mathrm{g}\mathrm{a}\mathrm{s}\right)\to {\mathrm{O}}_2\left(\mathrm{a}\mathrm{b}\mathrm{s}\right)$$

(1)

$$\mathrm{T}<150 °\mathrm{C}:{\mathrm{O}}_2\left(\mathrm{a}\mathrm{b}\mathrm{s}\right)+{\mathrm{e}}^{-}\to {\mathrm{O}}_2^{-}\left(\mathrm{a}\mathrm{b}\mathrm{s}\right)$$

(2)

$$150 °\mathrm{C}<\mathrm{T}<400 °\mathrm{C}:{\mathrm{O}}_2^{-}\left(\mathrm{a}\mathrm{b}\mathrm{s}\right)+{\mathrm{e}}^{-}\to 2{\mathrm{O}}^{-}\left(\mathrm{a}\mathrm{b}\mathrm{s}\right)$$

(3)

$$\mathrm{T}>400 °\mathrm{C}:{\mathrm{O}}^{-}\left(\mathrm{a}\mathrm{b}\mathrm{s}\right)+{\mathrm{e}}^{-}\to {\mathrm{O}}^{2-}\left(\mathrm{a}\mathrm{b}\mathrm{s}\right)$$

(4)

For n-type SnO₂, the gas–surface interaction process can be explained as follows (Figure 4) [85,86]. Firstly, O₂ in the air is adsorbed onto the oxygen adsorption site on the SnO₂ surface when heated to a specific temperature. Subsequently, oxygen molecules capture the conduction band electrons of MOS to form chemisorbed oxygen species (O₂⁻, O⁻, O²⁻) and simultaneously create an EDL at the MOS grain contact interface, which results in a higher barrier, increasing the resistance. When introducing a reducing gas such as ethanol, the gas reacts with the chemisorbed oxygen species and releases the electrons back into the MOS simultaneously, thus lowering the resistance. These oxygen species can desorb or be adsorbed again onto the MOS surface to create new oxygen anions. On the other hand, introducing an oxidizing gas will further deplete the sensing layer electrons and increases MOS resistance.

Although there is no unified theory on gas-sensing mechanisms, rational experimental designs based on existing materials can enhance their gas-sensing properties. The width of the space-charge layer, also known as the Debye length, typically ranges from 2 to 100 nm due to oxygen adsorption on the MOS surface. When the MOS grain size is close to or less than twice the Debye length, it amplifies the sensitivity considerably, demonstrating the remarkable potential of nanomaterials in gas sensors [91]. Nevertheless, excessively small particle sizes can cause nanoparticle aggregation, which will hinder the participation of internal particles in the reaction and decrease the gas-sensing properties of MOS [72]. Hence, specific nanoscale structural designs are necessary to reduce particle size while preventing nanoparticle aggregation. The heterojunction effect and synergistic effect can also impact the gas-sensing properties of MOS materials [92], which can be achieved through metal ion doping, metal particle modification, and compounding with other substances on the MOS substrate. In summary, the micro/nanostructure design and “second-phase modification” are essential methods for enhancing the gas-sensing properties of MOS materials. However, designing a new gas-sensitive material from scratch is still challenging.

2.2. Candidate Materials

Early comments on gas sensor design suggested that any sufficiently fine dispersed metal oxide could serve as a gas-sensitive layer for gas sensors, regardless of the practical usage requirements [93]. However, the exhaled gas of humans contains many components with low concentrations [30,32], making it necessary to identify the most appropriate candidate material among a wide range of available materials to accurately and efficiently measure the components and concentrations of exhaled breath [94]. We conducted a search using keywords such as “material name”, “chemical formula”, and “gas sensor” to investigate the most researched MOS over the past few decades. For the n-type semiconductors SnO₂, ZnO, TiO₂, WO₃, In₂O₃, Fe₂O₃, and CeO₂, the number of search results was 12,663, 10,920, 5269, 4102, 3487, 2643, and 1012, respectively, while for the p-type semiconductors CuO, NiO, Co₃O₄, Cr₂O₃, and Mn₃O₄, they were 3003, 2649, 1604, 822, and 564, respectively. The statistics show in Figure 5 that the overall proportions of the research results for the n-type and p-type MOS gas-sensitive materials were 82.3% and 17.7%, respectively. Among the top five materials, SnO₂, ZnO, TiO₂, WO₃, and In₂O₃ were n-type MOS materials, indicating that more attention has been paid to the research and development of n-type MOS gas-sensitive materials. According to Hübner et al., the inherent low sensitivity of p-type MOS materials restricts their development [95]. Furthermore, only TiO₂ belongs to the bulk resistance-controlled MOS among the top five materials. The conductivity change in them involves the reaction of gas with the lattice oxygen of the material, leading to a higher reaction temperature (usually above 700 °C), slower response rate, and poorer stability, making it less suitable for the detection of VOCs in exhaled breath [96,97,98,99].

According to the statistical analysis of exhaled breath biomarkers for lung cancer VOCs and MOS gas-sensing materials (Figure 3 and Figure 4), the n-type surface resistivity-controlled MOS materials have gained significant attention for detecting VOCs in exhaled breath for lung cancer diagnosis. Among them, SnO₂ and ZnO are currently the two most widely studied MOS gas-sensing materials and receive the most attention. Therefore, the following sections will focus on the latest research on n-type MOS gas-sensing materials (especially SnO₂ and ZnO), detecting exhaled breath biomarkers of lung cancer (BTEX, isoprene, hexanal, nonanal, 2-butanone, acetone, n-pentane, 1-propanol, etc.), considering the diverse types, low concentrations, and high-humidity conditions of exhaled breath analysis. The aim is to guide the development of e-nose devices with higher sensitivity, better selectivity, and more excellent stability to enable early lung cancer screening on a large scale.

2.3. Single MOS Gas Sensor and Performance Improvement Strategy

BTEX, comprising benzene, toluene, ethylbenzene, and xylene, has been frequently detected in the exhaled breath of lung cancer patients [56], as demonstrated by the statistical data presented in Figure 3. These compounds are metabolized similarly in lung cancer patients and are often detected together in exhaled breath samples [46,56]. However, due to the limited selectivity of MOS gas sensors for VOCs in this family, it is advisable to select one of the strongly correlated BTEX gases as a representative for detection [100]. For instance, the toluene content in the exhaled breath of lung cancer patients is approximately 2–3 times higher than that of healthy individuals, ranging from 80 to 100 ppb [101].

When evaluating the performance of MOS gas-sensing materials for detecting lung cancer exhaled breath biomarkers, several parameters should be considered beyond sensitivity and response time. In particular, the selectivity, limits of detection (LOD), and stability are significant. Selectivity refers to a gas sensor’s ability to detect the target gas in the presence of other gases, which is crucial, given that over 3000 VOCs and various inorganic gas matrices are in human exhaled breath [30,32,102]. LOD refers to the lowest target gas concentration corresponding to the gas sensor’s minimum reliable response sensitivity value. MOS sensors should exhibit high sensitivity to detect low-concentration (ppt levels) VOCs [25,103]. Stability means the ability of MOS sensors to obtain reliable results over time, which is essential for an accurate diagnosis [104].

2.3.1. Structures and Gas-Sensing Performance

The gas-sensing properties of MOS are significantly affected by its morphology and size [72,91]. A general strategy to enhance its performance is to optimize its structure and morphology, which can lead to a high specific surface area and excellent chemical activity [91,105]. SnO₂ is the most widely studied material in MOS gas sensors [106] due to its exceptional stability and electron transfer ability [107] (as shown in Figure 4). As a surface resistance-controlled MOS, SnO₂ provides abundant surface chemical properties owing to its bivalent states of tin elements (i.e., Sn⁴⁺ and Sn²⁺), making it more tunable than other univalent MOSs [108,109]. To date, various SnO₂ morphologies have been developed, such as nanoparticles [110], nanospheres [111], nanowires [112,113], nanotubes [114], nanofilms [115], nanosheets [116,117], and 3D nanostructures [118,119,120,121].

Figure 6a,b shows two schematic diagrams of the synthesis route of multi-layer SnO₂, respectively. Bing et al. successfully synthesized SnO₂ with a yolk-shell cuboctahedra porous structure formed by self-assembled nanoparticles (Figure 6a); the response to 20 ppm toluene was 28.6 at 250 °C, with short response/recovery times of 1.8 s and 4.1 s, respectively (Figure 6c) [120]. Compared with the solid cubes and single-shell structures of SnO₂ synthesized during the same period, the yolk-shell cuboctahedra structure of SnO₂ exhibited higher sensitivity and a shorter response/recovery time at the same concentration due to its loose and porous structure. This structure provides more active adsorption sites for gas–surface interactions and greater stability than the single-shell structure, thus promoting continuous gas–surface reactions. Wang et al. used a controllable multi-step synthesis method to obtain three-layer, double-layer, and single-layer hollow cubic SnO₂ samples at different experiment stages (Figure 6b). The gas-sensing test results for each at 20 ppm toluene at 250 °C showed that the responses were 38.7, 33.4, and 22.1, with corresponding response/recovery times of 0.8/6.1 s, 2.3/5.8 s, and 2.0/6.5 s, respectively (Figure 6d). It can be observed that the response of the sample enhanced with the number of layers in the hollow cubic structure. The three-layer hollow cubic structure of SnO₂ [113] exhibited a higher response value and a shorter response time compared to the “double-layer” yolk-shell cuboctahedra structured one [120] under similar testing conditions. This work demonstrates the significant promoting effect of multi-layer structures on the gas-sensing performance of SnO₂. However, the high LOD limits their application in detecting exhaled gas components in patients with lung cancer; thus, it is necessary to consider other methods to reduce the LOD.

2.3.2. Noble Metal Modification and Gas Sensing Performance

The surface modification of noble metals has proven to be an effective strategy for enhancing the response and selectivity of MOS gas sensors and reducing the operating temperature and response/recovery time [122]. It is worth noting that noble metal surface modification can be achieved through either chemical or electronic sensitization [79,123]. Chemical sensitization utilizes the catalytic effect of noble metal nanoparticles to directly boost the reaction rate between the target gas and chemisorbed oxygen [79]. In contrast, electronic sensitization indirectly enhances the gas sensor’s performance by transferring electrons from the MOS to noble metal, thus providing more oxygen adsorption sites [123]. For example, Qiao et al. reported achieving a LOD of toluene below 100 ppb by loading Pd nanoparticles onto a SnO₂ monolayer nanocage (Figure 7d,e) [119]. Moreover, the 1%Pd-SnO₂ monolayer nanocage demonstrated superior gas-sensing performance to a larger surface area of three-layered hollow cubic SnO₂, where the response of 20 ppm toluene was 41.4 and the response time was 0.4 s at 230 °C [121]. These results demonstrate the significant role of noble metal loading in improving the gas-sensing performance of MOS, which is attributed to Pd nanoparticles’ dramatic chemical and electronic sensitization properties (Figure 7a–e).

Suematsu et al. also investigated the potential of Pd-SnO₂ as a sensing layer in gas sensors, which has shown promising results [124]. They found that adding Pd to the original SnO₂ improved its catalytic activity and transformed its microstructure. Specifically, the addition of Pd converted densely packed SnO₂ nanoparticles (NPs) into SnO₂ nanoparticle clusters (CNPs), and Pd nanoparticles attached to the surface of SnO₂ CNPs to form Pd-SnO₂ CNPs (Figure 7f,h). The large pores (approximately 10 nm in radius) between the Pd-SnO₂ CNPs facilitate the adsorption and diffusion of toluene molecules inside, significantly shortening the response/recovery time (Figure 7g,i). The dual-action of Pd-loading significantly enhances the sensing performance of original SnO₂, allowing Pd-SnO₂ CNPs to respond well to toluene concentrations as low as 1 ppb. Furthermore, its LOD for toluene reaches 200 ppt, which far exceeds the detection limit requirements for exhaled breath detection. It is worth mentioning that they recently proposed modifying the heating pulse drive mode to improve the sensor response further, reducing the LOD of gas sensors for toluene to 7 ppt [125].

Selectivity is a crucial factor in assessing the performance of gas-sensitive materials. Numerous studies have demonstrated that incorporating noble metals can effectively enhance MOS’s selectivity [114,126,127,128,129,130]. For instance, Moon et al. designed a dual-layer gas sensor comprising a SnO₂ sensing layer and an xRh-TiO₂ (x = 0.5, 1, and 2 wt%) catalytic layer that solely catalyzes aromatic compounds, resulting in a selective response to BTX (benzene, toluene, and o-xylene) at the ppb level just by adjusting the Rh content in the Rh-TiO₂ catalytic layer [129]. The response of the pure SnO₂ sensing layer and three xRh-TiO₂ dual-layer gas sensors to the 5 ppm mixed gas (benzene, toluene, o-xylene, ethanol, formaldehyde, and carbon monoxide) is illustrated in Figure 7j–m. Within the operating temperature range of 325–425 °C, the xRh-TiO₂/SnO₂ sensors selectively responded to BTX compared to the pure SnO₂ sensors. The specific response behaviors of each sensor varied based on the Rh content, which demonstrates the possibility of customizing the selectivity of aromatic compound sensors by adjusting the Rh content of the catalytic layer. The quantification of BTX in the mixed gas was achieved successfully by forming the above sensors into a sensor array, as shown in Figure 7n.

In addition, the study of noble metals in multi-layer MOS gas sensors is not limited to the catalytic layer’s catalysis; it can also be used in the sensing layer. For example, Jeong et al. developed a CeO₂/Rh-SnO₂ dual-layer gas sensor in which CeO₂ acted as the catalytic layer, and Rh-SnO₂ acted as the sensing layer [130] (Figure 7q,r). They systematically investigated the effect of the type of MOS in the sensing layer (SnO_2, ZnO, In2O₃, and WO₃) and the type of noble metal loaded onto it (Pt, Rh, and Au) on the sensors’ selectivity and achieved quantitative differentiation of BTEXS (benzene, toluene, ethylbenzene, xylene, and styrene) at the ppb level [130] (Figure 7s,t). Unlike the Rh-TiO₂ catalytic layer of Rh-TiO₂/SnO₂ [129], the CeO₂ catalytic layer of CeO₂/Rh-SnO₂ [130] catalyzes interfering gases other than aromatic compounds, thereby avoiding the cross-response of the MOS sensing layer to interfering gases.

Figure 7. Noble metal modification and gas-sensing performance of SnO₂. Schematic energy band diagrams of a SnO₂ and Pd contact (a) with establishing the contact, with being exposed to (b) air and (c) toluene ambience, respectively. (d,e) Schematic diagrams of possible gas-sensing mechanisms of Pd-loaded SnO₂ porous cages. Schematic diagrams of the gas-diffusion behavior in (f) SnO₂ NPs and (h) Pd-SnO₂ CNPs. Transient response curve of the SnO₂ NPs microsensor to (g) 20 ppm toluene and (i) various toluene concentrations at an applied voltage of 1.04 V (~250 °C). Gas-sensing properties of (j) pure SnO₂, (k) 0.5Rh-TiO₂/SnO₂, (l) 1Rh-TiO₂/SnO₂, and (m) 2Rh-TiO₂/SnO₂ sensors to the 5 ppm mixed gases (benzene, toluene, p-xylene, ethanol, HCHO, and CO) within the operating temperature range of 325–425 °C. (n) PCA plot using the data from 0.5Rh-TiO₂/SnO₂, 1Rh-TiO₂/SnO₂, and 2Rh-TiO₂/SnO₂ sensors to demonstrate the discrimination of aromatic BTX compounds over the interferences from ethanol, HCHO, and CO (concentration: 1–5 ppm). (o) Schematic diagram of Rh-TiO₂/SnO₂ sensor. (p) Cross-sectional SEM image of Rh-TiO₂/SnO₂ sensing film. (q) Cross-sectional SEM and (r) FESEM images of CeO₂/Rh–SnO₂ film. (s,t) Normalized signal intensities of diverse single-layer sensors to 5 ppm analyte gases. Panels (a–e): reproduced with permission from Ref. [119], © 2016 Elsevier B.V. Panels (f–i): reproduced with permission from Ref. [124], © 2018 American Chemical Society. Panels (j–p): reproduced with permission from Ref. [129], © 2021 The Authors. Advanced Science published by Wiley-VCH GmbH. Panels (q–t): reproduced with permission from Ref. [130], © 2023, The Author(s).

Figure 7. Noble metal modification and gas-sensing performance of SnO₂. Schematic energy band diagrams of a SnO₂ and Pd contact (a) with establishing the contact, with being exposed to (b) air and (c) toluene ambience, respectively. (d,e) Schematic diagrams of possible gas-sensing mechanisms of Pd-loaded SnO₂ porous cages. Schematic diagrams of the gas-diffusion behavior in (f) SnO₂ NPs and (h) Pd-SnO₂ CNPs. Transient response curve of the SnO₂ NPs microsensor to (g) 20 ppm toluene and (i) various toluene concentrations at an applied voltage of 1.04 V (~250 °C). Gas-sensing properties of (j) pure SnO₂, (k) 0.5Rh-TiO₂/SnO₂, (l) 1Rh-TiO₂/SnO₂, and (m) 2Rh-TiO₂/SnO₂ sensors to the 5 ppm mixed gases (benzene, toluene, p-xylene, ethanol, HCHO, and CO) within the operating temperature range of 325–425 °C. (n) PCA plot using the data from 0.5Rh-TiO₂/SnO₂, 1Rh-TiO₂/SnO₂, and 2Rh-TiO₂/SnO₂ sensors to demonstrate the discrimination of aromatic BTX compounds over the interferences from ethanol, HCHO, and CO (concentration: 1–5 ppm). (o) Schematic diagram of Rh-TiO₂/SnO₂ sensor. (p) Cross-sectional SEM image of Rh-TiO₂/SnO₂ sensing film. (q) Cross-sectional SEM and (r) FESEM images of CeO₂/Rh–SnO₂ film. (s,t) Normalized signal intensities of diverse single-layer sensors to 5 ppm analyte gases. Panels (a–e): reproduced with permission from Ref. [119], © 2016 Elsevier B.V. Panels (f–i): reproduced with permission from Ref. [124], © 2018 American Chemical Society. Panels (j–p): reproduced with permission from Ref. [129], © 2021 The Authors. Advanced Science published by Wiley-VCH GmbH. Panels (q–t): reproduced with permission from Ref. [130], © 2023, The Author(s).

2.3.3. Improve the Humidity Resistance

The sensitivity of MOS gas sensors will decrease when operating in a high-humidity environment, resulting from “water poisoning” [131]. Such degradation is a major limitation to the reliability of gas sensors [104]. The primary mechanism involves the reaction of H₂O with the chemisorbed oxygen species on the MOS surface, forming inert hydroxyl groups at the oxygen-active site. This alteration significantly impacts the MOS gas-sensing properties [79,104,132,133,134] and gas diffusion [135]. Moreover, water vapor is an unavoidable and substantial interference gas for MOS in exhaled breath detection. Researchers have concentrated on improving the material’s humidity resistance to mitigate this issue.

The basic principle of eliminating the humidity’s effect is to suppress the formation of hydroxyls on the MOS surface or increase oxygen vacancies to mitigate the impact of hydroxyls. A practical approach is to avoid the connection between H₂O and the active sites on the MOS surface. Some researchers have used strong hydrophilics, such as CuO, to composite with MOS to block the interaction between H₂O and the MOS surface. CuO/SnO₂ [136] and CuO/In₂O₃ [137] have shown excellent humidity resistance (Figure 8a,b). Bulemo et al. reported a porous Pt-loaded SnO₂ nanotube based on SiO₂, which exhibited excellent sensing performance under high humidity (95% RH), attributing to the residual SiO₂ responsible for strong humidity adsorption [114]. Low-valence doping is another method typically carried out in MOS to provide oxygen vacancies to improve humidity resistance [131,138]. Kwak et al. doped Tb on the surface of yolk-shell sphere SnO₂, which demonstrated a comparable response and resistance in dry and humid environments (Figure 8c) [131]. This compensating effect can be attributed to substituting Tb³⁺ for Sn⁴⁺ in SnO₂, which creates oxygen vacancies. However, the humidity resistance capacity obtained by the additive materials is limited.

Another strategy for improving MOS’s humidity resistance is to enhance its hydrophobicity. Zhu et al. deposited a few nanometers-thick hydrophobic inorganic CeO₂ layer on SnO₂ thin film by magnetron sputtering to prepare the CeO₂/SnO₂ heterojunction film, making the material humidity resistant without being limited by other conditions (Figure 8d,e) [139]. However, the gas sensitivity of MOS is inevitably inhibited due to the hydrophobic layer covering its active sites. Jeong et al. developed a dual-layer sensor with a Tb₄O₇ coverage layer, which proved the universal humidity resistance effectiveness while maintaining a gas response, selectivity, and resistance (Figure 8f–k) [140]. The synthesis of surface-decorating materials requires additional materials and steps, increasing the complexity and cost of the process.

Microscopic morphology modification has also been proven to be an effective method for enhancing the humidity resistance of a single material [141]. Vallejos et al. fabricated ZnO thin films with rod- and needle-like structures (Figure 8l,m) [142]. Both ZnO structures show hydrophobicity, with static water contact angles (CA) of 120° and 134°, respectively, and maintain stable response values when only the humidity changes, indicating their excellent humidity resistance. The needle-like ZnO film with a larger contact angle has more excellent humidity resistance because it exposes more low-energy {100} facets, suggesting its lower reactivity to water. This specific morphology preparation method provides a novel approach to enhancing the humidity resistance of the material.

In addition to improving the humidity resistance of the materials themselves, gas pre-drying [143,144,145,146] and humidity compensation [147,148,149,150] can also enhance MOS gas sensor properties in high-humidity environments. Gas pre-drying can be achieved by evaporating [146] or condensing [143] the H₂O from mixed gases and can utilize hydrophilic sorbents, such as Nafion tubes [145], to absorb the H₂O from the sample before testing. Although gas pre-drying can alleviate humidity interference, it may remove the target VOCs, thus losing diagnostic information. An additional dehumidifying device will complicate the gas sensor and compromise its portability. Humidity compensation with reasonable algorithms [147,149,150] is also an effective strategy to reduce the hindering effects of humidity by incorporating a humidity sensor into the gas sensor system to measure the humidity levels and adjust the gas sensor’s output accordingly. Humidity resistance is pivotal in gas sensor design to ensure accurate and dependable performance in practical scenarios.

Figure 8. Improving the humidity resistance of MOS. Dynamic response and relationship between response and concentration under diverse humidity for (a) InCu₀ and (b) InCu₂. (c) Resistances of the pure SnO₂, 1Tb-SnO₂, 5Tb-SnO₂, and 15Tb-SnO₂ sensors under dry and 80% RH conditions. (d) Static water contact angle tests of SnO₂, CeO₂, and CeO₂/SnO₂ films. (e) Responses of SnO₂, CeO₂, and CeO₂/SnO₂ films at different RHs. Responses of pure (f) In₂O₃, (h) SnO₂, and (j) ZnO. Responses of (g) In₂O₃, (i) SnO₂, and (k) ZnO with Tb₄O₇ coverage layer. Static water contact angle tests of (l) rod-ZnO and (m) needle-ZnO. Panels (a,b): reproduced with permission from Ref. [137], © 2020 American Chemical Society. Panels (c): reproduced with permission from Ref. [131], © 2018 American Chemical Society. Panels (d,e): reproduced with permission from Ref. [139], © 2022 American Chemical Society. Panels (f–k): reproduced with permission from Ref. [140], © 2020 The Authors. Advanced Functional Materials published by Wiley-VCH GmbH. Panels (l–m): reproduced with permission from Ref. [142], © 2019 The Authors. Published by Elsevier B.V.

Figure 8. Improving the humidity resistance of MOS. Dynamic response and relationship between response and concentration under diverse humidity for (a) InCu₀ and (b) InCu₂. (c) Resistances of the pure SnO₂, 1Tb-SnO₂, 5Tb-SnO₂, and 15Tb-SnO₂ sensors under dry and 80% RH conditions. (d) Static water contact angle tests of SnO₂, CeO₂, and CeO₂/SnO₂ films. (e) Responses of SnO₂, CeO₂, and CeO₂/SnO₂ films at different RHs. Responses of pure (f) In₂O₃, (h) SnO₂, and (j) ZnO. Responses of (g) In₂O₃, (i) SnO₂, and (k) ZnO with Tb₄O₇ coverage layer. Static water contact angle tests of (l) rod-ZnO and (m) needle-ZnO. Panels (a,b): reproduced with permission from Ref. [137], © 2020 American Chemical Society. Panels (c): reproduced with permission from Ref. [131], © 2018 American Chemical Society. Panels (d,e): reproduced with permission from Ref. [139], © 2022 American Chemical Society. Panels (f–k): reproduced with permission from Ref. [140], © 2020 The Authors. Advanced Functional Materials published by Wiley-VCH GmbH. Panels (l–m): reproduced with permission from Ref. [142], © 2019 The Authors. Published by Elsevier B.V.

In summary, we reviewed how morphology modification and secondary modification can improve the performance of MOS gas sensors towards VOCs as lung cancer biomarkers in exhaled breath, considering the actual situation (diverse types, low concentrations, and high humidity) of exhaled breath analysis.

Table 1. MOSs have been reported for potential lung cancer biomarker VOCs.

Target Gas	Material and Structure	Concentration	Temperature (°C)	Response	Res/Rec Time (s)	LOD	Interference Gas	Ref.
Benzene	Co-Al2O3	5/10/50 ppm	100/100/100	1.66/2.53/21.86	1.95/2.18, 2.23/2.59, 2.87/3.15	-	-	[151]
Benzene	2Rh-TiO2/SnO2, dual- layer sensor	5 ppm	325	35	-	-	-	[129]
Toluene	WO3 mesoporous nanofibers	1 ppm	350	11	8.56/9.2	100 ppb	H2, H2S, CO, ethanol, NH3, CH4	[152]
Toluene	ln2O3	50 ppm	27	9	26/28	-	methanol, ethanol, acetone, n-butanol, and benzene	[153]
Toluene	WO3 porous nanostructure	100 ppm	225	132	2/6	-	methanol, acetone, glycol, formaldehyde, ethanol, C2H2, NH3, NO2, and CO	[154]
Toluene	Pd-SnO2 CNPs	1 ppb	250	3	-	200 ppt	O2, H2, N2	[124]
Toluene	Pd-SnO2 CNPs	7.9 ppb	250	15	-	7 ppt	Air	[125]
Toluene	1Rh-TiO2/SnO2, dual- layer	5 ppm	325	103	-	-	-	[129]
Toluene	SnO2@SnO2 yolk-shell cuboctahedra	20 ppm	250	28.6	1.8/4.1	-	benzene, methanol, acetone, and ethanol	[120]
Toluene	Three-, two-, one-layer hollow cubic SnO2	20 ppm	250	38.7, 33.4, 22.1	0.8/6.1, 2.3/5.8, 2.0/6.5	-	-	[121]
Toluene	Pd-loaded SnO2 cubic cages	20 ppm	230	41.4	0.4/16.5	100 ppb	-	[119]
Toluene	SnO2/NiO nanoparticle	100 ppm	250	66.2	-	10 ppb	-	[155]
Xylene	0.5Rh-TiO2/SnO2, dual-layer	5 ppm	325	120	-	-	-	[129]
Xylene	Pt/SnO2 nanosheet flowers	200 ppm	200	154	29/47	-	-	[156]
Xylene	Co3O4-SnO2 hollow nanostructures	5 ppm	275	18.6	243/-	-	ethanol, toluene	[157]
Xylene	CoWO4-Co3O4 heterojunctions composites	100 ppm	200	51.6	-	300 ppb	ethanol, methanol, formaldehyde, benzene, toluene, acetone, NH3, NO2, H2O	[158]
Styrene	Pt-SnO2/α-Fe2O3 hollow nanospheres	1 ppm	206	10.56	3/15	50 ppb	-	[159]
Isoprene	In2O3 nanoflowers	500 ppb	190	3.1	53/299	5 ppb	NH3, ethanol, H2, CO	[160]
Isoprene	ZnO quantum dots	1 ppm	350	42	42/8	10 ppb	-	[161]
Isoprene	In2O3/nanoparticles	1 ppm	350	231	3/35–200	1 ppb	acetone, H2, CO2, CO, CH4,	[162]
Isoprene	Pt-decorated In2O3/microspheres	5 ppm	200	103.5	124/204	5 ppb	H2O, CO, H2, ethanol, ammonia	[144]
Isoprene	1 wt%Cr2O3/In2O3 nanorods clusters	500 ppb	240	1.9	135/830	5 ppb	benzene, acetone, octane, pentane, ethanol, NH3, NO2	[163]
Hexanal	MnO2/Ti3C2Tx	20 ppm	100	52	134/381	-	-	[164]
Hexanal	In2O3 nanoparticle	50 ppm	300	18	-	-	-	[165]
Hexanal	CuO nanoflake	200 ppm	250	3.7	-	1.85 ppm	linalool, methyl salicylate	[166]
Hexanal	ZnO nanoparticle	5 ppm	250	2.12	-	-	1-pentanol, 1-octen-3-ol	[167]
Nonanal	Ru-W18049 urchin-like	30 ppm	RT	16.1	25/154	-	SO2, H2S, CO, NH3, ethanol, acetone,	[168]
Nonanal	Sb2WO6 hierarchical microspheres	30 ppm	RT	62	32/145	1.6 ppm	C8H16O, C9H14O, C6H12O, C10H18O	[169]
Nonanal	SnO2 nanosheets	0.1/0.3 ppm	250	1.383/2	-	-	CO, NO2, acetone, H2, ethanol, NH3, H2S, formaldehyde, acetaldehyde, butanal	[170]
Butanone	Ce-SnO2 cuboids	20 ppm	175	23.9	20/-	500 ppb	ethanol, toluene, acetone	[171]
Butanone	Pt-ZnO twin-rods	100 ppm	450	35.3	8/-	-	-	[172]
Butanone	Cr2O3/WO3 nanosheets	100 ppm	180	40.51	9/15	-	-	[173]
Butanone	1 at% Ce-SnO2 thin films	100 ppm	210	181	-	-	-	[174]
Butanone	ZnO small size	100 ppm	350	151	4.5/5	200 ppb	chlorobenzene, vinyl benzene, xylene, toluene, benzene, acetaldehyde, formaldehyde	[175]
Butanone	WO3 urchin-like mesoporous	50 ppm	240	188.5	7/13	100 ppb	-	[176]
Butanone	Ag-modified NiO porous spherical	100 ppb	320	3.2	5.5/8	50 ppb	Formaldehyde, methanol, acetone, acetaldehyde	[177]
Acetone	Ru-NiO flower-like microspheres	100 ppm	200	12	71/23	-	ethanol, methanol, formaldehyde, benzene	[178]
Acetone	TiO2/SnO2	100 ppm	300	301.5	-	20 ppb	ethanol, acetone, NO2	[179]
Acetone	PtCu-SnO2	5 ppm	240	27.8	-	5 ppb	ethanol, toulene, pentane	[180]
Acetone	Pt-ZnO-SnO2 porous nanofibers	100 ppm	170	104.26	-	-	C7H8, benzene, C3H6O	[181]
1-Propanol	Co-ZnO nanorods	100 ppm	250	491	2/19	10 ppb	formaldehyde, methyl alcohol, ethanol, triethylamine, 2-Propanol, benzene, ammonia, glacial acetic acid, formic acid	[182]
1-Propanol	ZnSnO3 nanospheres	10 ppm	200	10.3	10/90	500 ppb	acetone, xylene, ammonia, hydrogen, methane	[183]
1-Propanol	ZnO/NiO one-dimensional chain MOF	500 ppm	275	280.2	31.5/18.2	200 ppb	methanol, ethanol, isopropanol, hexanol, acetone	[184]
1-Propanol	PdO-ZnSnO3 hollow microspheres	100 ppm	140	30.8	1/25	-	formaldehyde, ethanol, acetone, xylene, methanol, ammonia	[185]
1-Propanol	ZnO nanoparticles	40 ppm	125	6.6	190/200	-	H2O, ethanol, acetone, benzene, toluene	[186]
1-Propanol	Cu2O double-shell hollow microspheres	100 ppm	187	11	50/40	10 ppm	acetone, carbon monoxide, ethyne, formaldehyde, isopropanol, ethanol, methanol	[187]
1-Propanol	NiO porous nanoparticles	20 ppb	75	1.59	-	20 ppb	ethanol, propanol, toluene, methane, NO2	[188]
2-Propanol	10 at% Co-ZnO nanoflower	5 ppm	225	22.5	330/475	/	N2, O2, CO2, acetaldehyde, isoprene, ethanol, acetone, methanol	[189]
2-Propanol	Fe-doped ZnO	250 ppb	275	4.7	51/762	250 ppb	H2O, ethanol, acetone, methanol	[190]

summarizes the latest achievements of some MOS gas-sensitive materials in detecting exhaled breath VOCs as lung cancer biomarkers.

2.4. Sensor Array and Pattern Recognition

This section focuses on developing the MOS gas sensor array for detecting potential lung cancer biomarkers in exhaled breath. To date, no single VOC has been identified as a specific biomarker for lung cancer [57]. Nevertheless, various highly sensitive MOS gas sensors have been developed to detect potential lung cancer biomarkers. However, detecting VOCs in exhaled breath poses significant challenges due to wide variations in composition [30], interference from other gases [32], and the low concentration of many biomarkers [25]. An ideal MOS sensor should be capable of detecting low concentrations of VOCs in the presence of high levels of interfering gases, such as water vapor, and respond rapidly to slight changes in concentration. It is difficult for a single MOS material sensor to meet all these requirements.

Inspired by biomimicry, Persaud et al. first proposed using electronic devices to mimic the olfactory system of animals in 1982, leading to the development of the e-nose [76]. The e-nose typically comprises an array of MOS gas sensors, signal acquisition, the preprocessing unit, and a pattern recognition algorithm [191]. Unlike the high selectivity of single MOS material sensors to a specific gas, each sensor in the MOS array need not have high selectivity for any given analyte. Instead, this approach records the responses between the exhaled breath and various MOSs, creating a set of specific signals known as “breath prints”. Based on the existing statistical models, the “breath prints” are analyzed to recognize a range of low-concentration lung cancer biomarkers’ VOCs [192]. As a simple example, Guntner et al. developed an array of five gas sensors (Figure 9a), which demonstrated excellent discrimination capabilities for mixed gases of ammonia, isoprene, and acetone [193]. Three sensors in the array were used for pattern recognition, and a simple multiple-linear-regression (MVLR) model was developed to analyze the linear response characteristics of analytes at sub-ppm concentrations (Figure 9b). Under 90% RH, the array achieved LODs of 2.9, 50.7, and 0.7 ppb for ammonia, isoprene, and acetone. These were cross-validated by the SRI-TOF-MS method, indicating the high reliability of the sensor array.

The sensor array effectively overcomes the limited selectivity of a single MOS sensor. A sensor array comprising multiple MOS gas sensors is generally required to analyze complex gases. However, using too many sensors can lead to information redundancy due to the MOS gas sensors’ broad-spectrum response characteristics, increasing the difficulty of recognition systems without improving accuracy [194]. Therefore, selecting the appropriate composition of MOS sensor arrays based on specific detection requirements [195] and developing more accurate and efficient pattern recognition algorithms are necessary [194,196,197]. Principal component analysis (PCA) and linear discriminant analysis (LDA) are two of the most commonly used dimensionality reduction algorithms to improve classification accuracy, reduce computational complexity, and facilitate the visualization of output results in gas recognition.

Figure 9c depicts the response of a sensor array comprising eight MOS sensors to five VOCs related to lung cancer biomarkers [198]. By combining the PCA method, the response of each sensor is portrayed as a feature vector with arrows in PC1-PC2, successfully identifying all gases (Figure 9d). Subsequently, testing was conducted in polluted air, and the target gases were identified accurately (Figure 9e). Furthermore, after removing three sensors, the array composed of fewer sensors showed a higher resolution for acetone and methyl isobutyl ketone when tested in the same polluted air (Figure 9f), emphasizing the importance of selecting suitable sensors [195]. Li et al. projected “breath prints” of lung cancer patients and healthy individuals onto the PC1-PC2 two-dimensional space [199]. They found significant overlap between the two groups (Figure 9g), indicating little difference in the main components of the exhaled breath between lung cancer patients and healthy individuals; then, using LDA to extract the features from the two groups, the categories showed a significant distinction between the two populations (Figure 9h). Combining the features extracted by PCA and LDA showed superior identification results (Figure 9i), demonstrating the advantages of different algorithms and their combinations. These results proved the applicability of the MOS gas sensor array in lung cancer diagnosis.

To summarize, MOS gas sensor arrays offer a promising technology for simultaneously detecting and identifying multiple gases. Environmental factors, such as temperature and humidity, can affect their accuracy and reliability, which need to explore new materials and fabrication techniques. In addition to PCA and LDA, many optimized algorithms have been introduced to improve gas recognition rates [200]. Machine learning and artificial intelligence techniques are also being used to improve gas identification algorithms.

Figure 9. Sensor array and pattern recognition. (a) Schematic diagram of the sensor array. Each sensor individually analyzes mixed gases, and a statistical model converts their signals. Utilize SRI-TOF-MS for training or cross-validation. (b) Average sensor array measurements of five volunteers’ ammonia, isoprene, and acetone concentrations as a function of entrapment time (the inside picture, RH, and CO₂ gas-sensing results). (c) Average sensor responses of eight sensors to 1 ppm of target gases in pure humid air. PCA scores and eigenvectors from (d) eight sensors in pure humid air, (e) eight sensors and (f) five sensors in polluted humid air. The 2D mapping results of all “breath prints” in features space, extracted by (g) PCA, (h) LDA, and (i) LDA with PCA. Panels (a,b): reproduced with permission from Ref. [193], © 2018 American Chemical Society. Panels (c–f) from Ref. [198]. Panels (g–i): reproduced with permission from Ref. [199], © 2020 Published by Elsevier Ltd.

Figure 9. Sensor array and pattern recognition. (a) Schematic diagram of the sensor array. Each sensor individually analyzes mixed gases, and a statistical model converts their signals. Utilize SRI-TOF-MS for training or cross-validation. (b) Average sensor array measurements of five volunteers’ ammonia, isoprene, and acetone concentrations as a function of entrapment time (the inside picture, RH, and CO₂ gas-sensing results). (c) Average sensor responses of eight sensors to 1 ppm of target gases in pure humid air. PCA scores and eigenvectors from (d) eight sensors in pure humid air, (e) eight sensors and (f) five sensors in polluted humid air. The 2D mapping results of all “breath prints” in features space, extracted by (g) PCA, (h) LDA, and (i) LDA with PCA. Panels (a,b): reproduced with permission from Ref. [193], © 2018 American Chemical Society. Panels (c–f) from Ref. [198]. Panels (g–i): reproduced with permission from Ref. [199], © 2020 Published by Elsevier Ltd.

3. Summary and Outlook

In recent decades, lung cancer, characterized by the highest mortality rate, has severely threatened human life and health. The primary reason for the high mortality rate is the late-stage diagnosis, emphasizing the importance of early screening. The use of exhaled breath analysis, a non-invasive, cost-effective, and user-friendly method, has been adopted to diagnose lung cancer. This article addresses the detection of VOCs in exhaled breath for early lung cancer screening using MOS gas sensors.

For a single MOS gas sensor, obtaining reasonable micro/nanostructure design and secondary modification can effectively enhance the gas sensitivity response by exposing more active sites on the MOS surface. Additionally, secondary modification is a proper technique for enhancing material selectivity. The humidity in exhaled breath can adversely affect the gas-sensing properties of MOS; therefore, this article summarizes ways to improve the humidity resistance of materials. Excellent humidity resistance composite materials can be obtained by compounding strongly hydrophilic or hydrophobic materials, low-cost doping to supplement oxygen vacancies, or exposing special crystal planes. Moreover, compared with modifying the material, pre-drying the gas before testing or using an algorithm for humidity compensation can also enhance the gas-sensing performance in a high-humidity environment.

Although the sensitivity of some single MOS gas sensors is exceptional, their broad response characteristics pose a challenge in detecting specific VOCs in exhaled breath. MOS sensor arrays can effectively address the problem of insufficient selectivity of single MOS sensors and are the future development direction of MOS gas sensing. For the analysis of more complex gases, sensor arrays comprising multiple MOS gas sensors are generally required, and efficient pattern recognition algorithms are critical to handling complex gas-sensitive information. The combination of deep learning technology and sensor arrays for detecting various diseases is an area that requires further exploration.

In summary, MOS gas sensors offer great potential for detecting human exhaled VOCs, which is significant for the early diagnosis of lung cancer on a large scale; however, several aspects require improvement:

• Clinical diagnosis. At present, the biomarkers of the exhaled breath of lung cancer patients have not been determined, which limits the application of MOS gas sensors in diagnosing lung cancer. We urgently need a single exhaled VOC, or a unified group of VOCs, as a standard marker for lung cancer to establish a highly reliable “breath prints” comparative database, which can significantly improve the accuracy of clinical diagnosis.
• Materials. The prerequisite for the pattern recognition of the sensor array is that MOS responds to low-concentration VOCs gas; therefore, the LOD of MOS needs to be further reduced. The high-humidity environment of exhaled breath and MOS’s high-working temperature seriously affect its stability and repeatability; thus, it is necessary to develop better humidity-resistant and lower working-temperature MOS materials.
• Algorithms. Deep learning algorithms based on olfactory recognition are needed to identify gases accurately in complex environments. Although still in its early stages, this technology has demonstrated strong recognition ability in other fields. Collaborating with sensor arrays is essential to achieve precise gas identification.
• Devices. The collaborative design and manufacturing of gas sensors using MEMS and CMOS technology reduces their size. Multiple sensors are integrated into a sensor array, and data processing modules enable chip-level packaging and manufacturing.
• Mechanisms. Understanding the gas-sensing mechanism involves complex chemical reactions, which are still not fully understood. Further research can improve the sensor’s performance, address selectivity, and stability issues, and guide the development of gas-sensing materials.

Through the collaborative efforts of multiple disciplines, it is possible to achieve breakthroughs in all the developments mentioned above and expedite electronic noses’ clinical deployment.