Real-time breath analysis for COPD risk assessment in smokers using a ZnO/SnO₂ heterojunction sensor integrated with support vector machine

Abstract

Chronic Obstructive Pulmonary Disease (COPD) is a progressively debilitating and life-threatening respiratory condition that has long been attributed to exposure to airborne toxic substances like carbon monoxide (CO), a reliable biomarker of oxidative stress, especially in smoking individuals. To address the need for a real-time and non-invasive method that can detect this life-threatening condition, a wearable diagnostic device is developed in this work based on a Z/S heterostructure thin-film gas sensor that can detect exhaled CO at a temperature as low as 37 °C. This gas sensor was prepared using a hydrothermal synthesis route and characterized by various techniques like X-ray diffraction (XRD), Field Emission Scanning Electron Microscopy (FESEM), and Energy-dispersive spectroscopy (EDS), and packed in a face mask design that enables continuous sampling of breath. Of all the sensing devices prepared with varying concentrations of the active components ZnO (Z), SnO₂ (S), and the combination of both (Z/S), the Z/S structure showed maximum sensitivity with a sensitivity of 264.29% at 12 parts per million (ppm), a response time of 14 seconds, and a recovery time of 3 seconds. The incorporation of a polymer layer of poly(3,4-ethylenedioxythiophene): poly(styrenesulfonate) (PEDOT: PSS) improved the charge carrier mobility through p-n junction formation. A customized gas sensing laboratory chamber enabled accurate calibrations of CO sensing. To derive actual CO levels from the resistance changes in the sensor, a power-law model was used. To classify patient groups further as current smokers, ex-smokers, and non-smokers accurately, a support vector machine (SVM) classification with a training accuracy of 94.2% and testing accuracy of 81.7% was used in this work. By integrating nanostructured gas sensors with wearable technology design concepts and ML diagnostics in a seamless Internet of Things (IoT) network, this scientific development will allow early diagnosis of the life-threatening condition of COPD in a completely non-invasive and non-destructive fashion with high portability. This developed device has unending possibilities in taking up roles in individualized breath medical care and represents a significant milestone in translating on-body disease detection using sensor technology.

Introduction

Chronic obstructive pulmonary disease (COPD), a progressive and irreversible respiratory disease, is characterized by symptoms of respiratory diseases, as well as limitations in air flow. It results in the death of over 3 million people annually, thereby making it the third leading cause of death globally¹. A staggering 2.9 million death rate and 45.4 million disability-adjusted life-years (DALYs) in people aged 70 years and older resulted from COPD in 2021 alone². It greatly affects the health of people everywhere, particularly the elderly. There are various environmental as well as behavioral risk factors that influence this situation. Biomass fuel use, smoking cigarettes, indoor and outdoor air pollution, as well as exposure to gaseous pollutants at work, all feature as environmental risk factors here³. Till date, despite various effective measures, the leading cause for COPD in developed countries, such as Norway, remains smoking cigarettes. This disorder presently affects 200,000 people in Norway, accounting for 6% to 7% of people aged 40 years and older here⁴. Statistics highlight the urgent need for continuous monitoring and early detection methods. CO has become one of the leading environmental pollutants, thereby playing an important role in the pathophysiology of COPD. This particular gas binds to hemoglobin thousands of folds when compared to oxygen, thereby forming carboxyhemoglobin that reduces the oxygen-carrying capacity of blood. In conjunction with oxidative stress, this reduces oxygenation, thereby accelerating the respiratory lesions in COPD patients here⁵. Notably, CO is not only a major constituent of cigarette smoke but also a byproduct of urban air pollution and biomass burning, making it a highly relevant biomarker for exposure assessment. Recent research emphasizes CO’s role in respiratory dysfunction. Here, the recent studies clearly make way for the role played by carbon monoxide in respiratory dysfunctions. In support, Yoshida (2024) clearly explained that higher levels of carbon monoxide result in the activation of cellular mechanisms, thereby leading to inflammation in lung tissue⁶. Furthermore, the study clearly indicated that those levels with higher levels of trace-risk components, such as CO, experience progression of the disease here. In similar lines, the study by Callejón-Leblic et al. ⁷ highlighted that levels of morbidity are well associated with greater levels of CO here. Evidence from occupation-related studies confirms this correlation, with Barbosa, J.V., et al.⁸ observing that CO and other gases exposed to firefighters led to a rise in the prevalence of asthma and COPD symptoms among these personnel over time⁸. These insights have driven innovation in sensing technologies, particularly for non-invasive respiratory monitoring. Chen, C et al.⁹ described a highly sensitive fiber optic sensor that employed a tin dioxide surface for detecting low concentrations of CO⁹. Others, Cuperus et al.¹⁰, investigated wearable technologies for patients suffering from COPD, encouraging the design of combined sensors that offer continuous respiratory information instantaneously¹⁰. Though welcoming, such sensing technologies are mostly at the prototyping level, meaning that these solutions are not yet fully integrated for practical applications in wearable forms, such as wearing masks. However, this similarity is yet to be fully harnessed, with proposals involving Metal Oxide Semiconductors (MOS) sensors, whose strong advantages include high sensitivity, selectivity, and rapid recovery rates. Z/S heterojunctions are particularly of interest, given the synergistic effect that these semiconductors present, wherein, at the junction material interface, the combination of S, which has higher electrical conductivity than other materials, paired with the larger surface area of Z, is particularly useful for increasing the process of charges transfer, or the detection, of gases¹¹. By facilitating effective electron transport, the n-n heterojunction that forms between Z and S improves the efficacy of CO detection. By incorporating these sensors into wearable face masks, real-time respiratory exposure monitoring is made easy and convenient¹². When it comes to differentiating between healthy people, smokers, ex-smokers, and patients with early-stage COPD, this method is quite helpful. Additionally, the integration of machine learning (ML) algorithms improves predictive modelling for disease progression and allows for the dynamic classification of user groups based on sensor output patterns¹³. Previous studies demonstrated that integrating ML classifiers with gas sensors based on nanomaterials greatly increases diagnostic precision. The current study suggests a wearable CO gas detecting device based on ZnO/SnO₂ nanostructures implanted in a face mask in light of these advancements. Exhaled CO concentrations are continuously monitored by the system, which also sends data for analysis. ML algorithms were used for classifying individuals according to their CO exposure profiles, specifically for identifying smokers, ex-smokers, and nonsmokers, and estimating their susceptibility to COPD. Furthermore, exhaled CO (eCO) has emerged as a robust, noninvasive marker for assessing respiratory dysfunction and oxidative stress in disease conditions. The concentration of eCO is known to remain in the range of 1-4 ppm for healthy individuals. However, according to their smoking status and disease progression, COPD patients always record elevated levels, specifically in the range of 2.9 ppm for non-smokers, 5.2 ppm for ex-smokers, and above 12.5 ppm for current smokers^14,15. A substantial rise in these levels takes place while progressing from Stages II (4.8 ppm) to Stages IV (8 ppm) in cases where the disease persists. This clearly indicates the applicability and usefulness of eCO measurement in real-time diagnostic systems and its possibility for incorporation in wearable health monitoring devices¹⁶. In Table 1, there is a comparative analysis of some novel CO-sensing materials, specifically based on Z-S composites, metal oxides (Cu, Pt, and In), polymer composites, and ternary nanocomposites, showing their sensing ranges, working temperatures, sensitivity, and dynamic responses, reportedly measured in earlier works on similar devices^17–23.

Table 1.

Overview of carbon monoxide sensing materials with performance metrics.

Sensing material	CO in (ppm)	Operating temp (°C)	Sensitivity (S)	Res. / Rec. Time (s)	Ref.
ZnO-SnO₂ Composite	5–2000	210-300	1.21 at 5 ppm	120/190-280	201617
Cu-Doped ZnO	100	95	60 %	--	202118
Pt-Doped SnO₂	100	Room Temperature	64.5	144/882	202119
Indium doped ZnO	1	300	1.84	--	202120
Polyaniline/(Ta₂O₅-SnO₂)	1	Room Temperature	--	--	202221
Zn₀.₉Mn₀.₁SnO₃	10–500	200	311.37 %	6.6 / 34.1	202222
ZnO-ZnWO₄ Composite	30	250	13x higher than ZnO alone	--	202423

Although there has been considerable progress made within CO detection technology using MOS, the important challenge that has remained unaddressed is its application within wearable devices that can be used for CO monitoring. This has remained a challenge despite materials being developed such as Cu/Pt doped MOS and photonics that can sense CO even at very low concentrations. However, these materials face important issues such as high operating temperatures. Moreover, while electronic nose (E-nose) systems based on Volatile Organic Compounds (VOC) analysis offer non-invasive screening, they frequently suffer from poor selectivity, high fabrication complexity, and interference from co-existing gases. Additionally, few systems provide dynamic classification of individual health states based on exhaled CO patterns, nor do they incorporate machine learning algorithms for intelligent disease risk prediction. Hence, there is a pressing need for a low-cost, room-temperature operable, and highly sensitive CO sensor that can be embedded into wearable platforms and combined with machine learning to enable personalized, real-time COPD monitoring. The literature analysis indicates that while several CO sensors, including those based on Z-S composites and in-doped Z, operate effectively, their use in portable or health-monitoring devices is restricted because they need high operating temperatures, usually above 200 °C. Even though Pt-doped S and other sensors function at room temperature, their usefulness for urgent detection requirements is diminished by their delayed response and recovery times. Even low CO concentrations can be dangerous for those with COPD; thus, sensors that can quickly and accurately detect trace levels of CO at low temperatures are desperately needed. The current work focuses on designing a material with enhanced responsiveness and high sensitivity requirements, motivated by the sensing capabilities of Pt- and Cu-doped metal oxides and Z-based composites documented in previous investigations. This study aims to fulfill the imperative requirement for early COPD diagnostic equipment, proposing the design of a wearable diagnostic device that incorporates a thin-film gas sensor with a Z/S heterojunction structure, integrated in a facemask diagnostic device, for real-time, non-invasive detection of the concentrations of CO, an effective biomarker that is found to increase significantly in concentrations in the breath of smokers and patients with COPD. This gas sensor works on the principle of detecting changes in the electrical resistance that varies when exposed to CO, which interacts with oxygen molecules adsorbable on the metal oxide surface to induce charges that cause changes in the metal oxide’s conductivity. In an effort to obtain quantified diagnostic accuracy, the metal oxide gas sensor has been designed to detect standard concentrations of CO ranging between 1 ppm to 20 ppm, reflecting the quantitative levels found within exhaled breath concentrations within various stages of COPD severity, as identified through scientific categorization. This non-invasive breath diagnostic system features an adjustable interface designed to direct exhaled breath through a tube to a closed metal oxide gas sensor chamber, designed to limit contamination of the detecting metal oxide layer while providing controlled exposure to the metal oxide gas sensor active layer. A microcontroller system supports real-time detection of changes in resistance, remotely wired to transmit cloud analytics through the ThingSpeak cloud computing system, while designed to categorize user’s respiratory pattern as either those of a COPD-risk-exposed individual, ex-smoker, or a non-smoker, applying an SVM-algorithm trained on models of resistance patterns of classified subjects, while achieving a normalized improvement of 33% enhancement in classification accuracy to 81.7% in testing, compared to a preceding accuracy of 61% before further enhancements, designed to classify COPD stages based on CO concentrations within the patient’s exhaled breath. This metal oxide gas sensor, intended for use in a biomedical setting, operates at a physiological temperature of 37 °C to enhance lasting, ergonomic, and continuous system use, while designed to feature a rapid response time of 14 s, 3 s recovery time, and overall sensitivity to concentrations of 264.29% at 12 ppm, reflecting system enhancement to achieve highest system efficacy, alongside system cost-effectiveness, scalability, and portability to enhance early-stage diagnostic effectiveness within at-risk COPD patient groups.

Experimental section

Materials and methods

In Table 2, the nature and quantity of starting materials, as well as the reaction time and temperatures used in the synthesis of the Z and S samples, have been summarized.

Table 2.

Summary of raw materials and synthesis parameters.

Sample name	Material	Quantity (m.mole)	Reaction time (Hrs.)	Temperature (°C)
Z	Zn(C₂H₃O₂)₂·2H₂O	0.5	6	120
S	SnCl₂·2H₂O	10	48	130

ZnO nanopowder preparation

The typical procedure for creating the Z, S, and Z/S sensors is shown in Fig. 1. It begins with dissolving 0.5 millimole (mmol) of of zinc acetate dihydrate and 3.6 gram (g) of urea in 80 milliliter (ml) of deionized water. For hydrothermal treatment, this solution is transferred to an autoclave lined with Teflon²⁴.

Fig. 1.

Step-by-Step process for synthesizing and fabricating Z, S, and Z/S sensors.

The product is allowed to return to room temperature following the reaction. Centrifugation is used to collect the blue precipitate, which is then extensively cleaned with ethanol and deionized water to get rid of any remaining reactants. The sample is then calcined for four hours at 350 °C to produce a pure ZnO nanomaterial.

SnO₂ nanopowder preparation

The stacked flower-like SnO₂ structure was synthesized through a hydrothermal method, as outlined in Fig. 1, starting with a solution of 10 mmol hydrated tin chloride and 1.4 g of urea in 100 ml of deionized water, mixed thoroughly. This solution was heated up for 48 hours (h) at 120 °C in an autoclave lined with Teflon. Centrifugation was used to purify the final product using ethanol and deionized water, and it was then left to dry overnight at 80 °C. The dry material was then crushed into a fine powder and heated for 6 h at 400 °C.

Fabrication of the sensitive Z/S heterojunction thin film

A 1 × 1 cm2 FTO substrate was taken and thoroughly cleaned using distilled water, ethanol, and acetone in an ultrasonic cleaner, followed by drying at 80 °C on a hot plate for 15 minutes (min) to remove volatile organic compounds. Concurrently, ZnO and SnO₂ nanopaste were prepared by grinding 0.5 g of the powder with 0.25 ml of ethanol" A further 0.5 ml of ethanol was added after 30 min, and the grinding process continued on for a further one hour. 0.5 ml of ethyl cellulose solution was gradually added at 30-min intervals, and then 0.25 ml of alpha-terpineol, until a homogenous paste was obtained. After mixing this paste with 0.5 ml of isopropyl alcohol (IPA), it was stirred for 2 h. After the substrates were cleaned, the resultant solution was coated using a doctor blade method and a spin coater set to 3000 rpm for 30 s and dried for 10 minutes at 60 °C. Three iterations of this coating and drying procedure were required to reach the required film thickness. Finally, the coated films underwent six cycles of heating at 150 °C for 10 min, followed by an hour of annealing at 350 °C, as shown in Fig. 1. Afterward, a silver paste was manually applied to the edges of the films to create electrical contacts, ensuring proper connectivity for sensor functionality.

Material characterization

A Cu-Kα radiation source with a wavelength of 1.5406 Å was used to record the synthesized materials’ XRD patterns. Using FESEM, the surface morphology of the ZnO/SnO₂ (Z/S) heterojunction nanomaterial modified by surfactants was examined. A source meter unit was also used to assess the fabricated devices’ current-voltage (J-V) behavior when exposed to light.

Results and discussions

X-Ray diffraction

The XRD patterns of Z, S, and the Z/S heterostructure are shown in Fig. 2. The hexagonal wurtzite structure is identified by the ZnO sample’s prominent peaks at 2θ values that correspond to the (100), (002), and (101) planes. Reflections from the (110), (101), and (211) planes are among the distinctive peaks that S displays that correspond to the tetragonal rutile phase. The heterostructure pattern shows clear peaks from S and Z, indicating that the Z/S composite with high crystallinity was successfully formed²⁵. The diffraction peaks for S align with the tetragonal phase as per Joint Committee on Powder Diffraction Standards (JCPDS) card no. 00-021-1250, while those for Z correspond to the hexagonal structure referenced in JCPDS 00-003-0888.

Fig. 2.

XRD patterns of Z, S, and the Z/S heterostructure samples.

The Debye-Scherrer equation (1) is used to estimate the average crystallite size of the nanoparticles²⁶. Table 3 presents the XRD results for samples Z, S, and Z/S, indicating the crystallographic planes, diffraction values, FWHM values, interplanar spacing, and corresponding values for the crystallite size, according to which the crystal structure properties of the sensing materials differ.

1

Table 3.

Crystallographic and structural parameters of synthesized sensor samples.

Sample name	Miller indices (h k l)	Angle θ (Radians)	FWHM β (Radians)	d-Spacing (nm)	Mean crystallite size (nm)
Z	(1 0 1)	0.3168	0.0036	0.0028	26.62
S	(1 1 0)	0.2331	0.0251	0.0193	3.96
Z/S	(1 0 1)	0.3164	0.0036	0.0027	19.66

Where λ is the wavelength of the X-rays (1.5406 Å for Cu-Kα), β is the full width at half maximum of the diffraction peak, θ is the Bragg angle, D is the crystallite size, and k is the shape factor (typically 0.9).

FESEM

Fig. 3(a) below gives the FESEM micrograph of the Z sample. Based on the micrograph, the surface of the sample has closely packed grains that have a relatively equal hexagonal shape²⁷. Such closely packed grains result in a continuous surface that favors the mobility of electrons, making this particularly desirable for gas-sensing devices.

Fig.3.

FESEM images of a) Z-sample, (b) S-Sample, (c) Z/S Sample, and (d) EDS.

In Fig. 3(b), morphology of the S sample has been shown. The particles are more irregular and have formed loose agglomerates, showing an inhomogeneous growth process²⁸. This inhomogeneous morphology could be because of the inhomogeneous synthesis process. These surface properties could provide a higher number of active sites for the interaction of the analytes. In Fig. 3(c), the composite of the Z/S sample has been shown. The morphology seems to be more fine and dense compared to the individual samples. The particles are more integrated and uniformly dispersed, showing a higher possibility of the formation of a heterostructure of Z and S²⁹. This dense morphology provides a higher charge-transfer rate along with a higher effective surface area, which are highly desirable for the improvement of gas sensor performance.

EDS analysis

EDS analysis was used to investigate the elemental composition of the Z/S composite, as seen in Fig. 3(d). It can be observed in the energy dispersive spectroscopy analysis that the Z/S composite is formed because the zinc (Zn), tin (Sn), and oxygen elements are present in the composite. These signals indicate the presence of these elements in the Z/S composite. However, a small peak of carbon is encountered in the analysis, which can be attributed to the sample support material. Most significantly, no other element is encountered, and thus it can be indicated that the Z/S composite possesses a high level of purity after being synthesized.

Energy band diagram of PEDOT: PSS/Z/S CO sensor

The PEDOT: PSS/Z/S heterojunction presents a p-n structure, and its energy band alignment is significantly affected by the exposure to CO gas. Interaction between CO and adsorbed oxygen species on the surface of Z and S results in electron release into their conduction bands, increasing electron concentration and narrowing the depletion width as shown in Fig. 4(a). This shifts the Fermi level upward in Z and S, hence promoting charge transfer across the junction. The HOMO and LUMO levels of PEDOT: PSS align well with the bands of Z and S, allowing efficient separation of carriers³⁰. Consequently, the overall conductivity increases and effective CO detection is enabled.

Fig. 4.

J–V characteristics of the fabricated devices: (a) Energy band Diagram of Z/S Sensor, (b) S (c) Z devices show nearly symmetric conduction, while (d) the Z/S heterojunction exhibits more linear and an ohmic like profile.

Current-voltage (J-V)

The electrical behavior of the fabricated heterojunction devices was investigated in an ambient condition by J–V measurements. Fig. 4(b–d) shows the current–voltage characteristics of the S, Z and Z/S configurations. Both the S and Z devices show roughly symmetric current flow upon forward and reverse bias, reflecting ohmic like transport dominated by interface conduction as opposed to pronounced rectification, which agrees with previous reports from PEDOT: PSS/Z junctions³¹. In contrast, the Z/S heterojunction presents a more linear J–V characteristic, in agreement with low barrier charge injection through the multilayer stack. The cut in voltage, or bias at which the forward current starts increasing more significantly, has been extracted for each configuration, amounting to 0.50 V for the S device, 0.48 V for the Z device, and 0.42 V for the Z/S device. This reduction of the cut in voltage in the Z/S heterojunction might be due to its benign type II band alignment, which gives rise to a built-in electric field at the interface. This internal field lowers the potential barrier, increasing electron transmission, therefore promoting efficient charge transport. In order to quantify the symmetry of current flow, the rectification ratio (R) has been calculated using the conventional definition (2)³²:

2

where I(+V) and I(−V) are the forward and reverse currents measured at the same magnitude of applied voltage.

The J–V data analysis (Table 4) has also verified that the rectification ratio for the three configurations remains nearly equal to 1, thus proving the symmetry of the current values for both forward and reversed polarizations.

Table 4.

Cut-in voltage, forward/reverse currents at ±0.90 V, and rectification Ratios For S, Z, and Z/S devices.

Device	Cut-in voltage (V)	I(+0.90 V) (A)	I(−0.90 V) (A)	Rectification ratio (R)
S device	0.50	0.0038	−0.0036	≈1
Z device	0.48	0.0039	−0.0037	≈1
Z/S device	0.42	0.0098	−0.0096	≈1

The difference between the three exists in the values of current efficiency rather than in the rectification properties, as the former (with the smallest voltage cutoff and higher values of the current under the same puncture voltages with a linear J-V relationship) for the Z/S heterojunction remains the best. These aspects are an outcome of the type-II band alignment properties, which ensure efficient injection and a low operation voltage. The data collectively reveal the better performance of the Z/S device compared to the S and Z structures and its usefulness for real-time breath analysis and low-power COPD risk estimation. Fig. 4(d) depicts the J-V curve for the Z/S heterojunction device. The linear current-voltage connection supports ohmic conduction, which is useful for sensing because breath detection relies on adsorption modulated variations in conductivity rather than diode rectification. A higher current at reduced bias, accompanied by stable linear conduction, demonstrates the Z/S system’s capability for real-time breath analysis and low-power COPD risk assessment.

Gas sensing chamber design

To evaluate the Z/S heterojunction thin-film sensor’s ability to detect different CO levels, a specially made CO gas sensing chamber was created. As shown in Fig. 5, the configuration had two distinct gas inlets, one giving CO and the other nitrogen (N₂).

Fig.5.

Schematic diagram of a gas sensing system.

To ensure precise control over the gas mixture’s concentration, mass flow controllers were used to govern both gases. To safely release the extra gases, a bubbler was attached to the outlet. To guarantee constant gas exposure, the sensor was placed on a holder inside the chamber. A heater coil was placed inside the chamber and kept at 37 °C with the use of a temperature controller in order to replicate human body conditions. A PICOTEST M3510A multimeter was used to record the sensor’s resistance readings in real time. An integrated computer system was used to continuously capture and analyze the data. This setup offered a dependable setting for assessing the sensor’s reaction to various CO concentrations in regulated and repeatable circumstances³³.

Gas sensing mechanism

In air (Baseline Condition):

When the sensor is exposed to air without the CO target gas, oxygen molecules are adsorbed onto the Z/S layer’s surface. Free electrons from the metal oxide materials’ conduction band are captured by these adsorbed oxygen species as (3).

3

By reducing the quantity of free electrons in the conduction band, this interaction creates a depletion layer, which is shown in the picture as a blue area. This layer raises the resistance of the sensor.

Upon exposure to target gas (CO)

The target gas, like CO, interacts with the deposited oxygen ions when it comes into contact with the sensor surface. The oxygen species’ previously detained electrons are released by this reaction and return to the metal oxide’s conduction band. For example, when CO is involved, the reaction can be shown as (4).

4

Since the depletion layer at the sensor surface narrows as a result of this reaction, the concentration of free electrons rises and the resistance of the sensor falls. In general, the resistance in air (R_air) is greater than the resistance when the target gas (R_gas) is present. For the detection of reducing gases, this change in resistance serves as the basic sensing mechanism.

Heterojunction role (Z/S) and PEDOT: PSS coating

When ZnO and SnO₂ are integrated, an n-n-type heterojunction is created, creating an internal potential barrier at their interface. This barrier increases the mobility and separation of charge carriers, which raises the material’s total gas-sensing efficiency.

Additionally, by coating the heterostructure with PEDOT:PSS, a p-type conductive polymer, the n-type metal oxides and the p-type polymer interact to generate a p-n junction. As seen in Fig. 6, this p-n junction greatly increases the sensitivity of the sensor and improves charge transport across the interface. The connection permits a more noticeable change in carrier concentration when exposed to the target gas, which immediately results in a stronger and more sensitive sensing signal³⁴.

Fig. 6.

Gas sensing mechanism.

Response/recovery time study

Here, a comparative analysis of the gas-sensing behavior of PEDOT: PSS-coated S, Z, and Z/S heterojunction sensors was conducted.

A higher electron mobility and surface reactivity of SnO₂ allowed more efficient CO gas adsorption in PEDOT: PSS/S sensors (27-7 s). As a result of the abundance of active sites on the SnO₂ surface and the improved charge transfer dynamics enabled by the conducting polymer, this configuration was notably more sensitive than the ZnO sensor. The PEDOT: PSS/Z sensor, on the other hand, exhibited a moderate response time/recovery time (30-11 s), attributed to its wide bandgap and relatively balanced surface adsorption parameters. A moderate recovery time was also observed, since the PEDOT: PSS layer influenced gas molecule desorption by slightly hindering CO gas release. A PEDOT: PSS-coated Z/S heterojunction sensor was found to have the fastest response time and the highest sensitivity, as shown in Table 5. To achieve this improved performance, n-n heterojunctions are formed, which enhance charge separation and accelerate surface reactions with target gases. As a result of deeper trap states at the hetero interface and fast desorption kinetics caused by the polymeric coating, this sensor also exhibited a relatively short recovery time (14-3 s). The comparative performance is depicted in Fig. 7, which shows the response and recovery profiles of (a) the S-sensor, (b) the Z-sensor, and (c) the Z/S sensor under identical test conditions. The sharp and unsaturating part in Fig. 7(a) confirms previous observations for SnO₂ gas sensors, for which the prolongation of adsorption times and conductivity changes due to oxygen vacancies impede the sensor from saturating. This also happened for PtOx/SnO₂/a-C gas sensors exposed to CO, as shown in the work of Kim et al. in 2024³⁵, where the resistance again did not saturate during gas exposure.

Table 5.

Performance parameters of the proposed co sensors.

Proposed sensor	Response time (S)	Recovery time (S)	Sensitivity (% S, 12 ppm)
Z	30	11	25.86
S	27	7	165.71
Z/S	14	3	264.29

Fig. 7.

Response and recovery time study of (a) S-sensor, (b) Z-sensor, and (c) Z/S sensor.

Sensitivity

The sensitivity of a gas sensor refers to its ability to detect and respond to changes in its concentration³⁶. In resistive-type sensors, this is typically measured by observing the change in electrical resistance when the sensor is exposed to the target gas compared to its baseline resistance in clean air (5).

5

Where R_a is the resistance of a sensor in clean air conditions. R_g is the resistance of the sensor of a target gas.

The elevated sensitivity observed at low CO concentrations can be explained by the plentiful adsorption sites present on the sensor surface, which cause a more significant relative change in resistance. It should be noted that as the concentration of CO increases, the above active sites are filled progressively, causing a decrease in the value of the resistance change and thus the sensitivity percentage, as demonstrated in Fig. 8. This phenomenon occurs especially in the Z/S sensor. The synergistic effect occurring at the heterojunction interface worsens this scenario.

Fig. 8.

(%) Sensitivity of Z, S, and Z/S sensors to changing concentrations of CO gas ranging between 4 to 20 ppm.

Resistance – CO concentration calibration with error representation

Results from calibrations reveal that there is a clear and detectable resistance change corresponding with the increase in CO concentration for the Z/S sensor designed and developed in this work. The resistance change is ideal for quantitative measurement of CO concentration, as illustrated in R² = 0.998, indicating that there is a significant correspondence between measured data points and their corresponding equation on the graph³⁷. The standard error of estimate, at 1.38 KΩ, further indicates that all data points strongly cluster around the equation line, indicating reliability in measurement. A significant linear association was evident on resistance change and CO concentration, as illustrated in Fig. 9 and compiled in Table 6.

Fig. 9.

Calibration of resistance vs. CO concentration. Circles correspond to the experimental data. The solid line corresponds to a linear regression fit (R=3.677⋅PPM+16.387, R²=0.998). Shaded regions are corresponding to 95 % CI of the mean (dark band) and 95 % PI for individual values (lighter band). The scatter of data around the fit is represented by the standard error of estimate and equals to 1.38 KΩ.

Table 6.

Calibration data of resistance versus co concentration with linear regression fit.

CO Concentration (ppm)	Resistance (KΩ, measured)	Regression fit (KΩ)	Findings
4	30	31.1	Within 95 % CI/PI
6	40	38.4	Within 95 % CI/PI
8	45	44.3	Within 95 % CI/PI
12	60	60.5	Within 95 % CI/PI
16	75	74.3	Within 95 % CI/PI
20	90	89.0	Within 95 % CI/PI

This predictable linear response, since the concentration values are readily derived from resistance, is useful for real-time monitoring and makes integration into electronic readout systems easier. Potential application in wearable or portable diagnostic platforms is supported by the sensor’s constant PI and narrow CI, which further emphasize the reproducibility of performance.

Application of a PEDOT: PSS-Coated Z/S CO gas sensor for assessing COPD severity

COPD is a progressive, life-threatening non-communicable respiratory disorder for which no curative treatment exists at present. It mainly presents itself in the forms of emphysema, characterized by the destruction of alveolar air sacs, and chronic bronchitis, related to inflammation of the lining of the bronchial tube. In the last couple of years, medically approved self-testing kits have really revolutionized disease detection by way of easier and faster diagnoses. Of these, E-nose systems have emerged as promising tools for non-invasive disease identification through the analysis of exhaled breath. Such systems generally include a gas sensor, a data processing unit, and a machine learning model for pattern recognition. A tailored PEDOT: PSS-coated Z/S CO gas sensor is used as the primary sensing element in this investigation. It is designed to detect changes in the amount of carbon monoxide in exhaled breath, an essential biomarker associated with the severity of COPD. The sensor provides a platform for real-time, accessible, and compact respiratory health monitoring when combined with microcontroller-based hardware and the right algorithms. This helps with early-stage identification and better treatment of COPD. Owing to the sensor’s adaptability, its primary features complement eCO’s clinical significance as a vital COPD biomarker. Insights about oxidative stress and respiratory inflammation in patients can be collected by detecting higher eCO levels. This further enhances the use of the PEDOT: PSS-coated Z/S sensor for personalized diagnostics. Through the integration of clinical information with the output from the sensor, it is possible to make an accurate diagnosis concerning the severity of COPD. This further improves the use of the PEDOT: PSS-coated Z/S sensor for personalized diagnostics. Furthermore, it is now possible to make an accurate diagnosis concerning the severity of COPD. This further supports pro-active disease monitoring and management.

Correlation between exhaled CO and COPD severity

In support of this, the use of eCO has increasingly been acknowledged in the scientific community to be a non-invasive marker of respiratory exposure or oxidative stress. It has been proved that in healthy non-smokers, the range of eCO was between 1 and 4 ppm, but in COPD patients, the value was higher. Non-smokers with COPD ranged about 2.9 ppm, ex-smokers ranged about 5.2 ppm, and among current smokers, it was over 12.5 ppm¹⁴. It was concluded that smokers had significantly higher levels of eCO, which measured an average of 9.69 ± 3.11 ppm compared to 2.19 ± 0.98 ppm for non-smokers. The difference was found to be significant (p-value < 0.001). Among the smokers, the value was even higher in COPD patients (mean value of 10.45 ± 3.03 ppm) in contrast to others (7.05 ± 1.56 ppm) but was highly significant (p-value < 0.001). However, even higher increases in the level of eCO have also been measured while progressing to COPD, which measured 9.54 ± 3.32 ppm in mild COPD patients, 10.44 ± 2.43 ppm in moderate patients, and 11.32 ± 3.15 ppm in severe patients³⁸.

System integration and patient monitoring setup

Fig. 10(a) depicts the final design, in which a face mask is connected to a pipe that directs a breath sample into a sealed container containing the sensors. Such a contained atmosphere is beneficial in ensuring that the breath samples collected are of excellent quality and not polluted by outdoor air. To increase safety, all of the sensors and other components are connected to the Arduino UNO via a wooden box.

Fig. 10.

(a) Top view Photographic image of CO monitoring Setup, (b) CO absorption response of the SnO₂-based sensor (4–20 ppm), sampled at 0.5 Hz for 500 s using a 10-bit Arduino Uno ADC. Processed data were baseline-corrected, low-pass filtered (0.2 Hz), and calibrated with a second-order polynomial fit (R² > 0.99).

System design and implementation

The proposed CO sensing system integrates an SnO₂-based sensing element with a custom-built signal conditioning circuit, microcontroller unit, and wireless communication interface, as illustrated in Fig. 10(b).

The core signal interface is an R-to-I (resistance-to-current) converter, where the sensing element (R_CO) changes resistance in response to CO concentration. This resistance variation alters the current in the op-amp feedback loop. A 741 operational amplifier converts the current variation into a proportional voltage drop across a precision load resistor (Rₛ = 1 KΩ), generating the primary analog signal. The voltage across Rₛ is connected to the Arduino Uno’s analog input pin (A0), with the op-amp and Arduino grounds tied together for a common reference. The Arduino’s onboard 10-bit ADC samples this voltage at 0.5 Hz (one sample every 2 s) for a total of 500 s per measurement cycle, producing 250 readings. This acquisition rate was selected after tests at 5 Hz confirmed that the slow sensor dynamics (tens to hundreds of seconds) are fully captured at the lower rate. Prior to digitization, the signal passes through an RC low-pass filter (~0.2 Hz cutoff) to suppress noise. The Arduino also interfaces with a DHT11 sensor for temperature and humidity monitoring and a micro SD card (SD) module for local data storage. Recorded voltage data are baseline-corrected, digitally low-pass filtered, and analyzed to extract key parameters such as peak response, rise time, recovery time, and percentage sensitivity change. Calibration is performed in a controlled CO chamber for concentrations ranging from 4 to 20 ppm, with the resulting second-order polynomial fit achieving an R2 > 0.99. It ensures a consistent level of performance. In the context of remote monitoring, an ESP-01 Wi-Fi module assists in transferring the processed information to a smart phone or a distant server, which has been symbolically depicted in Fig. 10(b). Overall, the sensing and acquisition process consumes about 22 milliwatts of power and provides a total end-to-end latency of about 5 ms.

Dynamic resistance variation in response to exhaled CO level

The resistance pattern of the CO gas sensor was distinguishable between normal and COPD-smoker responses, as evident in Fig. 11. The resistance of normal persons was relatively low, ranging between 100 and 450 Ω, as shown in Fig. 11(a). On the other hand, COPD patients were associated with high resistance values, ranging from 50 to 250 KΩ, as evident from Fig. 11(b). This was due to variation between gaseous composition and reaction with the Z/S layer of the gas sensor. Additionally, while normal persons showed minimal resistance changes, high resistance variation was measured for COPD patients. Moreover, the resistance measured for current, former, and non-smokers was normally distributed, as evident from Fig. 11(b). This was associated with the non-linear response of metal-oxide semiconductor gas sensors, which was related to gas concentration. Therefore, power law was appropriate for resistance-capacity analyses.

Fig. 11.

Resistance Variations (a) Healthy volunteers, (b) COPD patients.

Role of humidity background and CO-induced surface modulation

Human exhaled breath typically contains a large amount of humidity, even close to the saturation point (approx. 95-100% relative humidity) at the physiological temperatures of 34-37°C. Their concentration of water vapor is a few orders of magnitude higher than that of the trace gas biomarker carbon monoxide, with typical concentration levels in the range of 1-20 ppm. Consequently, the dominant role of humidity, in contrast to resistance modulation, is a fixed ambient condition in breath analysis. In metal oxide semiconductor sensors, the main contribution of humidity is related to the baseline conductivity, a consequence of surface hydroxylation and proton migration. Nonetheless, the reaction between reducing gases such as CO and surface-chemisorbed oxygen ions is selectively pronounced even in a humid environment. The depletion layer is affected by the released free charge carriers, that is, the electrons, returning from traps to the conduction band, thus providing a distinctive contribution to the resistance signal change according to ^39,40. Emittance of 1-4 ppm, characteristic of healthy persons, hardly influences the surface equilibrium state of oxygen, ideally providing a weak contribution, that is, a shallow variation in the resistance signal. Elevated levels of this trace gas, routinely exceeding 8-15 ppm, have been recognized in the case of COPD patients as well as in smokers according to⁴¹. Their higher concentration profoundly affects the rate of the surface reduction reactions, enhancing the modulation rate of the charge carriers’ density, hence the resistance signal. Hence, the main contribution in the resistance difference between COPD patients and healthy persons is related to the surface reaction rate coefficients solidly accompanied by a fixed humid atmosphere, instead of variations in ambient humidity. According to experimental findings from breath-analysis investigations, the reliable detection of trace gas concentration with the aid of metal oxide semiconductor sensors, even under the near-saturated humidity ambient, does not hamper the detection accuracy.

Breath sample cohort: demographic profile and classification based on CO levels

The study involved the collection of breath samples from 15 adult participants who were divided into three groups: (i) non-smokers, (ii) smokers, and (iii) Ex-smokers. The self-reported history served as the basis for classification, which was confirmed by the fabricated Z/S heterojunction sensor measuring breath CO levels. The volunteers, who ranged in age from 20 to 45, were in good health. Table 7 provides a summary of the CO concentration ranges, classification criteria, and demographic data. Under protocol number BME/ECC/2025/00-11, dated 18.06.2025, the Institutional Ethics Committee (IEC) granted ethical permission for the study. All human subject procedures were carried out in compliance with the Declaration of Helsinki. Before a breath sample was taken, each subject gave their written informed consent.

Table 7.

Participant demographics and classification.

Group	No. of participants	Age range	Gender (M/F)	CO Level range (ppm)	Classification criteria
Non-smokers	5	20–35	2M / 3F	<5	No history + CO <5 ppm
Smokers	5	25–40	4M / 1F	12–18	Self-reported + CO >10 ppm
Ex-smokers	5	30–45	3M / 2F	5–10	Quit >1 month ago
Total	15

Selectivity considerations in human breath analysis

Human breath contains a complex mixture of water vapor, carbon dioxide, carbon monoxide, as well as trace amounts of volatile organic compounds (VOCs) like acetone, ethanol, and isoprene. In the case of metal-oxide semiconductor gas sensors, the selectivity of the sensor for a single gas species or type can be considered impractical or impossible, whereas a selectively sensitive property can be conferred by the choice of semiconducting materials, operating temperatures, or their surface reaction rates. In the current experimental analysis, the Z/S heterojunction gas sensor has been employed that works at room temperature⁴², where carbon monoxide has a beneficial surface reaction kinetic rate with chemisorbed oxygen species, whereas the concentration of regular breath VOCs requires a higher operational temperature for surface oxidation. It has been previously indicated that the surface oxygen ions can be effectively reduced with the activation energy that is lower for carbon monoxide, as opposed to larger VOC molecules, showing decreased breath gas adsorption and desorption rate characteristics within a normal breathable regime with a sluggish activation rate at room temperatures⁴³. The Z/S semiconducting heterojunction arrangement can favorably support increased space charge solute interaction characteristics within the sensing layer, including the involvement of reducing substances like carbon monoxide. The occurrence of the semiconducting Z/S heterojunction within the sensing mechanism can stimulate a pronounced concentration sensitivity towards the diffusion characteristics of smaller breath molecules, whereas suppressing the overall influence or role of regular diffusive VOC molecules, acknowledging their decreased concentration interaction within the sensing layer. Similar experimental arrangement arrangements have previously shown significant selectively sensitive characteristics within humid as well as multi-gas detection toward carbon monoxide in a breathing process. In the context of the current experimental analysis, the concentration of carbon monoxide has been shown as a dominant amplifying factor toward the observational sensing variations, whereas the influence of the regular concentration variations of diffusive VOC molecules does not show a similar significant amplification. According to a physiological aspect, the concentration of carbon monoxide within the breathing process of COPD patients, as well as smokers, appears significantly higher than that related to normal healthy persons, whereas the concentration of the remaining diffusive breath VOC molecules appears much lower that does not show similar prominent significant amplifications^44,45.

CO concentration estimation via power-law modeling

To estimate CO concentration based on sensor readings, a power-law model was applied, describing the dependence of gas levels on sensor resistance (6).

6

Where R denotes the measured sensor resistance (in Ω), C represents the CO concentration in ppm, and and A and n are empirical constants obtained through calibration with known CO values. This modeling technique aligns with the findings of Yan et al.⁴⁶, who demonstrated that MOS sensors exhibit a power-law relationship between resistance and CO concentration. Fig. 12(a) illustrates the steps in the power-law modelling process, and Fig. 12(b) depicts the power-law curve that was fitted using the data that was collected.

Fig.12.

(a) Flow of power-law model, (b) Fitted power-law curve with observed data points.

Software deployed

The software implementation of the system uses the ThingSpeak platform, which is an open-source IoT platform that supports real-time data acquisition, cloud storage, and analysis. The ThingSpeak platform facilitates easier communication between IoT and cloud environments and ensures remote access and continuous observation. In addition to that, advanced analytics are made possible through embedded MATLAB code. The choice of the SVM algorithm in the proposed system for classification purposes considers its accuracy and capability to handle high-dimensional complex data. This supervised learning technique is excellent at determining the best boundaries between classes, which makes it a good choice for differentiating between people with and without COPD. The process, shown in Fig. 13(a), starts with the use of a sensor-integrated mask to take breath samples from both healthy people and COPD patients. The SVM model is trained using the collected data after it has been combined and preprocessed to identify patterns indicative of COPD. The individual breathes into the mask while the device is operating in real time, and an Arduino and Wi-Fi module send the sensor data to ThingSpeak. On the basis of these known patterns, the trained system analyzes this real-time information to ascertain if a person is a COPD patient or a healthy one.

Fig. 13.

(a) Flow chart, (b) Correlation map.

The correlation analysis, as evident from Fig. 13(b), shows that the outcome variable is most negatively related to the ex-smoker group (-0.27), followed by non-smoker (-0.24), and finally by the current smoker group (-0.14). The negative correlation in this context depicts that patients in these smoking groups are more likely to have COPD, given that the smoking variables are functioning as binary indicators (1 indicates that a person belongs to that group). Within this correlation analysis, the outcome variable is COPD, in which a person is marked as "1" if diagnosed with COPD, while a healthy person is marked as "0." This correlation map is preceded by a description within Table 8, according to which COPD diagnosis in previous studies has essentially been based on E-nose systems incorporating sensor arrays that identify a combination of several VOCs present in exhaled breath samples from COPD patients and healthy individuals combined. Not with standing their noninvasive nature, aforementioned studies have often faced issues like a lack of specificity, concurrent patterns in COPD and other conditions, along with a difficulty in easier analysis stemming from VOC patterns’ broad nature in a combined form. The novel study, in this regard, presents a Z/S sensing system that zeroes in on CO, a gas that is most strongly associated with smoking-induced oxidative stress in COPD patients and healthy ones combined. Utilizing a power-law model within a combination of this sensing system and a present-day SVM system, this novel study presents a more specific, faster, and more viable approach to point out COPD in a preliminary COPD test setup.

Table 8.

Performance comparison with the existing methodologies.

Material	Method	Findings	Limitations	Ref.
MOS Sensor Array	SVM on VOCs	Noninvasive Detection of COPD and Lung Cancer	Cross-sensitivity, low specificity to CO	47
Custom E-Nose System	VOC analysis	Discrimination of COPD and Lung Cancer	Lack of reproducibility	48
CNT Sensor Array	Clustering + Classification	Detecting COPD through breath analysis	Expensive fabrication, less accessible	49
E-Nose + SVM	Breath-print recognition	Identification of Breath-Prints for COPD Detection	Overlap of signals with lung cancer	50
ZnO/SnO₂ E-Nose	Direct CO detection using Power-Law + SVM	Targeted CO measurement, Real-time signal conversion, Optimized for smokers	Still focused on CO as a primary indicator; may need integration with other VOCs for broader profiling	This work

Conclusion

The current work proposes a holistic method of early diagnosis of Chronic Obstructive Pulmonary Disease (COPD) using the design and fabrication of a wearable non-invasive exhaled breath analyzer using a ZnO/SnO₂ (Z/S) thin-film resistive sensor with a heterojunction structure. The nanocomposite material was designed using the hydrothermal method, the structure, morphology, and composition of which was analyzed using the X-ray Diffraction (XRD), Field Emission Electron Scanning Microscope (FESEM), and Energy Dispersive X-ray Spectroscopy (EDS). The existence of separate crystal structures and homogeneously spread material elements signified the successful creation of the high-purity junction material.

The prepared sensor showed improved electrical properties, as indicated by the J-V analysis, where the Z/S structure showed the lowest cut-in voltage (0.42 V) compared to other variants, signifying effective charge carrier transport at the n-n heterojunction interface. Gas-sensing performance assessment showed that the Z/S sensor showed superior performance characteristics, such as a high sensitivity value of 264.29% at 12 ppm concentration, short response time of 14 s, and short recovery time of 3 s, with excellent repeatability characteristics under repeated exposure cycles. This clearly supports the use of the sensor for real-time analysis of trace concentrations of CO in a physiological setup.

To facilitate the implementation, the sensor was integrated with a mask platform, which was connected to an Arduino-powered embedded system. This made it possible to read resistance data and transmit the information wirelessly in real time to a cloud platform, ThingSpeak, using a Wi-Fi communication module. To this end, a gas sensing chamber was designed to allow precise delivery of breath samples to the sensing interface. Using power law modeling, the resistance change was processed to determine the actual concentrations of CO.

In order to further develop the system towards smart health diagnostics for COPD, SVMM was trained on a dataset consisting of exhaled CO values for both current smokers, former smokers, and non-smokers. Training and test accuracies reached 94.2% and 81.7%, respectively. In fact, it is important to highlight here that what makes the proposed system different and better than other existing Electronic Nose technology-based systems is its ability to provide CO detection, fast dynamic response, operating temperature of 37 °C, and support for machine learning for autonomous risk prediction.

The proposed system is an exemplary instance of the confluence of various disciplines such as nanomaterial-based gas sensing technology, embedded systems development, cloud-based IoT infrastructure development, and AI. It has the potential to become an effective and affordable solution at the confluence of point-of-care pulmonary diagnostics and telemedicine applications. It adds value to the existing pool of literature by providing an innovative means of real-time COPD risk assessment and has good potential in the context of providing preventive pulmonary healthcare.

Author contributions

Poundoss Chellamuthu, Kirubaveni Savarimuthu, Gulam Nabi Alsath M: Conceptualization, Methodology, Software, Visualization, Investigation, Writing- Original draft preparation. R. Krishnamoorthy, T. Yuvaraj, Mohit Bajaj: Data curation, Validation, Supervision, Resources, Writing - Review & Editing. Mohammad Shabaz: Project administration, Supervision, Resources, Writing - Review & Editing.

Data availability

The complete dataset used to support the conclusions of this research has been submitted to Zenodo, where it is openly accessible at: https://doi.org/10.5281/zenodo.15655842.

Declarations

Competing interest

The authors declare no competing interests.

Ethics

The study was approved by the Institutional Ethics Committee, Department of Biomedical Engineering, [Chennai Institute of Technology, Chennai-600069, Tamilnadu, India], under protocol number BME/ECC/2025/00-11, dated 18 June 2025.

All research procedures were performed in line with applicable national regulations and the ethical principles described in the Indian Council of Medical Research (ICMR) guidelines for biomedical and health research involving human participants, as well as the Declaration of Helsinki.