Comparison of Sampling Systems for Biological Sample Dehumidification Prior to Electronic Nose Analysis

Abstract

It is well known that gas sensor responses are affected by the presence of humidity in the analyzed gas. This is particularly true when dealing with biological fluid samples, whose high moisture content interferes with the adsorption of the trace volatile organic compounds (VOCs) on the sensors’ active layer. To address this challenge, this study focuses on designing and testing a novel sampling system for the dehumidification of biological fluid headspace to be characterized by an electronic nose (e-Nose). Such a system, based on the use of disposable polymeric sampling bags purged with dry air, exploits the polymers’ permeability to water vapor to reduce sample humidity. Tested materials included Nalophan^TM (20 μm), high-density polyethylene (HDPE, 8, 9, 10 and 11 μm), low-density polyethylene (LDPE, 12 and 50 μm), and biodegradable polyester (Bio-PS, 15 μm). First, dehumidification performance was characterized as a function of dry air flow rate and film type. A purge of 1 L/min accelerated the sample humidity removal compared to passive storage of bags from >2 h to <1 h (from 80% to 20% RH). Second, a mass-balance model was applied to dedicated experiments to decouple water losses due to diffusion and adsorption, showing that diffusion through the polymer wall dominates, while adsorption occurs in the early stages of conditioning. Third, because these materials are not selectively permeable to water, potential loss of water-soluble VOCs during dehumidification was investigated. Pooled urine headspace samples—both raw and spiked with a metabolite mix of VOCs—were dried using each material and analyzed using a photo-ionization detector (PID) and an e-Nose. Results were compared against a Nafion^TM dryer. Comparison was based on the e-Nose’s ability to discriminate between pooled vs. spiked samples and reveal real-life metabolomic changes. Nalophan^TM bags and Nafion^TM dryer provided the highest VOC fingerprint to support discrimination by the e-Nose, while Bio-PS provided the fastest sample dehumidification. The proposed bag-based system offers a cost-effective, disposable, and contamination-free solution to humidity interference in e-Noses.

1. Introduction

Inspired by the human olfactory system, electronic noses (e-Noses) combine an array of gas sensors and pattern recognition to detect and discriminate odors [1]. These systems are becoming of interest in several fields of application, such as environmental monitoring [2,3,4], food quality assessment [5,6,7], process control, and disease detection for healthcare [8,9], because of their versatility and relatively low cost. Despite promising results, several challenges still need to be solved before their deployment in real-life applications. These include poor sensor response stability due to drift over time, scarce reproducibility, and cross-sensitivity to environmental factors, which affect sensors’ responses by altering the kinetics and thermodynamics of the chemical reaction that occurs between the gas and the sensor’s active layer [1,10,11].

Among the interfering factors, humidity is one of the most critical issues [12,13,14]. High moisture content in a sample or fluctuations in ambient humidity can drift the sensor baseline and decrease sensor sensitivity, as water molecules compete with volatile organic compounds (VOCs) in surface reactions that determine the response of the sensors [15]. This effect of sensitivity reduction is particularly pronounced for samples with high inherent humidity, and may become critical when analyzing breath or headspace of bodily fluids, where high humidity levels (such samples are mostly water-saturated) are combined with very low concentrations of biomarkers of interest, which are often in the sub-ppm range [10,16].

Over the years, different strategies have been explored to mitigate humidity interference in gas sensing. Software-based approaches apply algorithmic compensation techniques by building empirical models that search for linear or non-linear relationships between humidity content within the gaseous samples and sensor responses [13,17] or based on machine learning models (e.g., ANN and PLS) to correct sensor outputs using humidity measurements [10,18,19]. Hardware strategies, by contrast, focus on material engineering by increasing the hydrophobicity of the sensing probes and the substrates to mitigate humidity interference [20,21] and proper design of the sampling system to physically remove or reduce sample humidity before analysis [15]. For these reasons, the latter approaches are preferred, especially for saturated biological samples, because removing sample moisture can improve sensor sensitivity. For example, pre-concentration methods (e.g., sorbent tubes, cryogenic traps, and solid phase micro-extraction) [22,23,24,25] and dehumidification techniques (e.g., condensers and solid desiccants) can effectively remove humidity directly on the sample gas line [26]. However, desiccants often introduce unavoidable and undesired risks of cross-contamination, while condensers may add complexity and cost to the sensing setup, which is problematic for portable or low-cost applications [15].

Another strategy to remove humidity from gas samples exploits the permeability to water vapor of polymeric materials [27,28]. For instance, Nafion^TM membranes are widely used for this purpose thanks to their strong hydrophilicity and high permeability to water vapor, which makes them effective for continuous moisture removal from gas streams [29,30]. However, Nafion-based dryers are relatively expensive, and when reused between samples, carry a risk of cross-contamination [26,31]. Such aspects restrict their practicality in biomedical applications, especially where contamination-free systems are desired [32]. Polymeric sampling bags made of other materials have also been used to reduce humidity via passive diffusive permeation. Nalophan^TM bags, commonly used for olfactometric analysis (EN 13725:2022 [33]), are known to be highly prone to the diffusion of small hydrophilic molecules (e.g., water, ammonia, or hydrogen sulphide) [34]. Previous work exploited Nalophan^TM capability to reduce the moisture content of urine samples by using the water vapor concentration gradient in ambient air as a driving force [35]. In applications involving biological fluids, polymeric bags are advantageous for their simplicity, low cost, and disposability, which avoids risks of cross-contamination between samples. Despite these advantages, polymeric bags are not selective for water vapor; some VOCs, especially low-molecular-weight or water-soluble compounds, may also diffuse out, leading to analyte loss and changes in the sample’s chemical fingerprint during storage [36,37]. Moreover, this approach typically requires long conditioning times (2 to 24 h in ambient air) to reach equilibrium, making it impractical for medical devices that require carrying out several analyses in one day.

In this context, the present research proposes an improved sample-preparation method that accelerates the dehumidification of gas samples during storage by combining polymeric permeation with an active dry air purge. By flushing the exterior of a polymeric sample bag with a controlled flow of dry air, the humidity gradient across the bag wall increases, accelerating the diffusion of water vapor out of the sample. This approach aims to achieve faster moisture removal than passive conditioning while retaining the advantages of using low-cost, disposable bags to prevent cross-contamination. The effect of dry air flow rate on dehumidification time and the performance of the proposed drying system was evaluated in a first phase using different polymeric materials, including Nalophan^TM, high-density polyethylene (HDPE), low-density polyethylene (LDPE), and biodegradable polyester (Bio-PS) (Phase 1). As humidity loss through polymeric films can occur via two mechanisms—diffusion through the film driven by concentration gradients, and adsorption of water onto the wall of the polymer [38,39]—a mathematical model was developed to quantify humidity losses resulting from the individual contributions of each mechanism under a no-purge configuration (Phase 2).

An important consideration is that none of the tested polymer materials are selectively permeable to water, and co-permeation of water-soluble volatile compounds is expected during sample dehumidification. Therefore, the study further evaluated whether the proposed drying method affects the presence of target VOCs and the e-Nose’s ability to discriminate among samples. To this end, we carried out experiments on human urine headspace—a biologically relevant sample matrix with a complex metabolomic profile. Urine was selected as the test matrix because it can be easily collected and stored (frozen) for long periods and contains VOCs with a broad range of polarities and chemical complexity, making it particularly suitable for biomarker discovery, clinical diagnosis, and treatment monitoring [28,40]. This preliminary study aimed at an initial screening of various materials; to that end, tests were conducted using a simplified experimental procedure with pure urine versus spiked urine, without any real reference to a real-world application. Using a combination of photoionization detector (PID) measurements and e-Nose sensor analysis, we compared urine samples prepared with the proposed drying systems based on polymeric bags against those dried with a conventional Nafion^TM dryer, which served as the dehumidification unit in previous reference work [41], but turned out to be prone to cross-contamination between samples. This constituted Phase 3 of the study. The impact of each drying method on VOCs content and on the discrimination of sample classes was assessed to verify that effective humidity reduction can be achieved without sacrificing critical diagnostic information. The overarching goal is to improve the reliability of e-Nose measurements in high-humidity biological samples by removing humidity interference while preserving the chemical information essential for detection and classification.

2. Design of the Experiment

2.1. Experimental Setup

The experimental setup consists of three parts: the sampling system, the drying system, and the e-Nose analysis system (Figure 1). The sampling and drying systems have two parallel lines: one for the sample and one for the reference air. The sample line is the part of the sampling system in which the gas sample (relative humidity (RH) ≈ 100%) is generated by bubbling air through a liquid sample. The reference line produces humid air (RH ≈ 100%) by bubbling air through distilled water, serving as reference air for the e-Nose sensors’ measurements. The entire system is installed inside a climatic chamber (HPPeco, Memmert GmbH, Schwabach, Germany) kept at 60 °C and 5% RH, as previously described by [35]. The whole system is supplied with compressed, filtered, dry air at 20–25 °C and 6–10% RH. Mass flow controllers (MFCs, MC5SLPM, Alicat Scientific Inc., Tucson, AZ, USA) are used to control the airflow through the bubblers and to set the VOC stripping degree. Both lines are operated at user-defined flow rates and durations, set within the MFC operational range.

While the sampling system arrangement is the same for the reference line and the sample line, two different working principles are implemented for the drying system. For the reference line, we utilized a Perma Pure^TM gas dryer MD-070-48P-4 (Lakewood, NJ, USA), housing a single Nafion^TM tube (120 cm length, 0.070 in outside diameter), to remove humidity from the humid reference air. This Nafion^TM tube was enclosed in a 1/4 in PTFE/PVDF tube that was continuously purged with a counterflow of compressed air controlled by an MFC. This creates a gradient in the water vapor pressure across the Nafion^TM membrane, driving the transfer of water molecules from the humid reference air inside the tube to the purge-gas flow outside of the tube [42]. Such a membrane has already been demonstrated to be effective in reducing the humidity content of biological samples, but it is prone to cross-contamination since it is in direct contact with samples [41,43]. In this work, the outlet of the Nafion^TM tube was connected to the inlet of a 2 L Nalophan^TM bag to collect the reference air before analysis. A temperature and humidity sensor (SHT40, Sensirion, Zurich, Switzerland) is placed inside the bag to measure humidity continuously. This system yields a reference air at approximately 20% RH at 60 °C, which was used as the target humidity level to be achieved during dehumidification of the humid gas sample in the sample line.

For the sample line, we implemented a novel bag-based drying system in place of a Nafion^TM dryer. A 2 L disposable sampling bag made of the selected polymer is placed in a Plexiglass/Teflon chamber (24 cm × 20 cm × 24 cm; volume: 11.52 L). Once the bag is filled with a humid gas sample, the solenoid valve V1 (2V025-08, Heschen, Foshan, China) is closed to physically separate the sampling bag from the bubbler and prevent backflow of gas. Dry air (RH < 10%) is then flushed into the drying chamber, surrounding the bag. This creates a high-water vapor concentration gradient between the humid gas samples inside the bag and the chamber environment, greatly accelerating the diffusion of water out of the bag’s wall. The chamber serves as a closed system with a uniform temperature and humidity distribution, isolated from external environmental disturbances. A SHT40 sensor is placed inside the bag to measure the decrease in sample humidity over time. When the sample humidity reaches the reference air level (RH ≈ 20%), the drying process is interrupted by turning off the chamber purge flow, and the dried gas sample is ready for analysis.

The measurement system is a laboratory-scale e-Nose prototype developed at Politecnico di Milano. It consists of a 300 mL stainless-steel chamber equipped with an array of seven metal oxide (MOX) gas sensors from Figaro Engineering Inc. (Osaka, Japan) (i.e., 3 × TGS2611, TGS2600, TGS2602, TGS2603, and TGS2620) along with a humidity and temperature sensor (SHT40, Sensirion, Zurich, Switzerland). A piezoelectric pump (MZB1001T02, Murata Electronics, Cernusco, Italy) is used with a custom-made flow sensor to maintain a constant flow of 300 mL/min in the chamber. Three solenoid valves (2V025-08, Heschen, Foshan, China) are used to select the e-Nose input, automatically switching between ambient air (V4), sample (V2), and reference air (V3) during analysis. During sample preparation, V2 and V3 are closed, while V4 is kept open to maintain continuous airflow in the chamber (see Section Urine Headspace Analysis for timing details). A microcontroller (NUCLEO-F746ZG, ST Microelectronics, Agrate, Italy) is used to control the MFCs, solenoid valves, pump, and data acquisition. A graphical user interface (GUI) developed in LabVIEW^® 2025 Q3 is used to control the different phases and save data.

2.2. Sampling Bags

Eight polymer-based materials commonly used for food packaging were selected to fabricate sampling bags (about 2 L) to serve as a drying medium in the proposed drying system. The tested materials and thicknesses included: Nalophan^TM (PET, Tillmanns S.p.A, Milan, Italy, film thickness 20 μm), high-density polyethylene (HDPE, Soderplast, Rho, Italy, film thickness 8, 9, 10, and 11 μm), low-density polyethylene (LDPE, Soderplast, Rho, Italy, film thickness 12 and 50 μm), and biodegradable polyester (Bio-PS, Soderplast, Rho, Italy, film thickness 15 μm) which is a blend of Poly(lactic acid) (PLA) and Poly(butylene adipate-co-terephthalate) (PBAT). All the materials were purchased as film rolls, and the sampling bags were produced in-house. Two PTFE tubes (6 mm i.d.) were fitted on each side of the bag for gas inlet and outlet. A visualization of all the bags is provided in the Supplementary Materials (Figure S1). These materials are widely used in food packaging because their water-vapor barrier properties [44]. In this study, it was assumed that the potential release of interfering volatile compounds from the bag materials remained constant, acting as a background for the e-Nose analyses.

2.3. Biological Samples Involved for Testing

A primary goal of this study was to determine whether drying samples with polymeric bags results in any loss of VOCs relevant to disease detection. To that end, we simulated a disease-classification scenario by preparing two classes of urine samples: pure pooled urine and pooled urine spiked with VOC biomarkers.

2.3.1. Urine Collection from Volunteers

Urine samples were collected at the Laboratory Olfactometry for the study from 12 volunteers aged 20–40 years, including 5 females and 7 males. First-morning samples from volunteers who fasted for 8 h were preferred over spot urine samples, given their reliability, as recommended by [35]. All volunteers provided informed consent, and the study was approved by the Ethical Committee at IRCCS Humanitas Research Hospital (Approval no. CE-ICH260/11). Samples were collected in sterile, individually packaged cups.

2.3.2. Pooled and Spiked Urine Sample Preparation

To introduce biological variability, three independent batches of urine mixtures were prepared. Each batch consisted of pooled urine from approximately four donors of both sexes. For the classification tasks, each batch was homogenized and divided into two equal portions:

• The first portion, labeled Class P, consisted of unmodified pooled urine samples.
• The second portion, labeled Class A, consisted of pooled samples spiked with 0.156 µL/L of 4-heptanone and 3.22 µL/L of acetone as specific chemical markers. Acetone and 4-heptanone were chosen based on a previous study suggesting their association with prostate cancer-related metabolic pathways [28,45]. The added concentration falls within the range of concentrations naturally found in human urine, as previously described [46].

The pooled samples were homogenized using a magnetic stirrer, aliquoted into 20 mL portions (sufficient for a single analysis), and stored at −18 °C to preserve their composition until further use. On the day of the analysis, the aliquot was thawed in a water bath and preheated to approximately 50 °C. Subsequently, the sample was transferred into the bubbler for headspace generation. This approach allowed direct evaluation of the e-Nose’s ability to discriminate between pure and chemically enriched samples. Experimental procedure has been defined based on previous research [35] and is illustrated in the Supplementary Materials (Figure S5).

2.4. Experimental Protocol

2.4.1. Phase 1: Dehumidification Performance of Sampling Bags

Dry Air Flow Rate Optimization

To preliminarily assess the effect of the dry air purge flow rate—hypothesized to have a limited impact beyond a certain point—dedicated experiments were performed using two representative materials (HDPE 11 μm and LDPE 12 μm). In these tests, a 2 L sampling bag made of the polymer film was filled with a humid gas sample generated by bubbling compressed air (135 mL/min) through 20 mL of distilled water in the sample bubbler for 15 min. With the filled bag placed inside the drying chamber, three dry air purge flow rates (1, 5, 10 L/min) were applied, and the results were compared with a static condition (i.e., 5% RH at 60 °C with no purge flow). For each condition, the dehumidification time—defined as the time required for the RH inside the filled bag to fall within 5% RH of the reference-air value (~20% RH at 60 °C)—was recorded. Analyses were done in triplicate.

Dehumidification Time Characterization of Sampling Bags

After determining the optimal dry air purge flow to be used for the screening experiments, the dehumidification performance of sampling bags made from all candidates was evaluated. This included Nalophan^TM, Bio-PS, HDPE (8, 9, 10, and 11 μm), and LDPE (12 and 50 μm). We compared the humidity profile measured inside each bag during drying of a humid air sample using the proposed bag-based drying system with the reference approach (i.e., Nafion^TM dryer). In each test, a humid air sample was generated by bubbling 20 mL of distilled water with 135 mL/min air for 15 min. For the bag-based system, the sample was then dried by placing it in a chamber and flushing the bag’s exterior with 1 L/min of dry air—selected based on the results of Section 3.1.1. For the reference approach, the humid outlet from the bubbler was connected directly to a Nafion^TM dryer to dehumidify the stream in real time. The Nafion^TM dryer was operated with a dry countercurrent flow rate of 300 mL/min. The dryer output was collected in a 2 L Nalophan^TM bag to serve as a reference sample. In both cases, an RH&T sensor inside each bag continuously measured the humidity decay over time. Regardless of its initial RH values (refer to Section 3.1.2), the dry purge stopped when the RH value measured in the sample bag reached a fixed value of 20% RH. Experiments for each bag type were run in triplicate.

2.4.2. Phase 2: Investigation of Adsorption and Diffusion Losses

Experimental Protocol

To investigate the individual contributions of water diffusion and adsorption during the dehumidification process using the sampling bags, the approach proposed by Eusebio et al. [38], entailing the modeling of H₂S loss through Nalophan^TM bags, was used as a reference. Authors proposed increasing the available surface area inside the bag by inserting an additional Nalophan™ film, thereby enhancing compound loss due to adsorption. To do this, only the sample line of the setup shown in Figure 1 was used, with no chamber purge, so that we could observe the natural diffusion and adsorption processes in the bag. Before each test, an empty sampling bag was placed inside the chamber at 60 °C and 5% RH for 12 h to stabilize the initial conditions. After this conditioning phase, the bag was rapidly filled with humid air by bubbling dry air at 2 L/min through 20 mL of distilled water, filling the 2 L bag in 1 min. This fast filling minimizes any significant diffusion or adsorption during the filling process. After filling, the bag was sealed and left in the chamber to undergo natural dehumidification, allowing humidity to diffuse out of the polymer film into the dry external environment and to adsorb onto the polymer’s inner walls (or dissolve into the polymer matrix). RH&T data were continuously recorded inside the bag until the equilibrium with the chamber conditions was reached. The experimental setup is illustrated in Figure S2 of the Supplementary Materials.

Six materials were considered for the test: Bio-PS (15 μm), Nalophan^TM (20 μm), HDPE (8 μm and 11 μm), and LDPE (12 μm and 50 μm). Each material was tested under a second configuration, using bags with the same geometric characteristics (i.e., a volume of 2 L and a surface area of 1380 cm²). In this second configuration, a sheet of film, made from the same polymeric material and having the same surface area as the internal surface of the 2 L bag, was placed inside the bag. The purpose of adding the polymeric sheet into otherwise identical bags was to increase the available surface area for water adsorption, thereby isolating the contribution of water vapor adsorption within the bag material itself. Tests for each material and configuration were performed in triplicate.

Calculation of Water Mass Loss

To quantify the relative contributions of adsorption and diffusion to total water vapor losses, a mass-balance-based mathematical model was developed using data from two independent experiments (bag with and without an internal film). First, RH (accuracy of ±1.8% RH) and temperature (accuracy of ±0.2 °C) measured inside the bags were converted into water mass (g) using the ideal gas law (Equation (1)), considering the pressure value depending on the temperature (Equation (2)).

$$\mathrm{m}=\left({\displaystyle \frac{\mathrm{P}\times \mathrm{R}\mathrm{H}}{100}}\right)\times {\mathrm{M}\mathrm{W}}_{{\mathrm{H}}_2\mathrm{O}}\times {\displaystyle \frac{{\mathrm{V}}_{\mathrm{b}\mathrm{a}\mathrm{g}}}{\mathrm{R}\times \mathrm{T}}}$$

(1)

$$\mathrm{P}=\exp \left(65.81-{\displaystyle \frac{7066.27}{\mathrm{T}}}-5.976\times \ln \left(\mathrm{T}\right)\right)$$

(2)

where P is the saturation vapor pressure in Pa at the temperature inside the bag (60 °C), ${\mathrm{M}\mathrm{W}}_{{\mathrm{H}}_2\mathrm{O}}$ is the molecular weight of water, V_bag is the bag volume in m³, R is the universal gas constant in J/(mol.K), and T is the temperature in °C.

The RH values recorded and corresponding water mass calculated during the experiments are reported in the Supplementary Materials (Figures S3 and S4). It is important to account for the bag volume throughout the experiment. At the initial boundary condition (t = 0), the bag volume is equal to zero, and it increases linearly during the filling phase at a flow rate of 2 L/min. The inlet gas is assumed to be fully saturated at 100% RH and 60 °C. This linear volume increase continues until the bag reaches its 2 L capacity after 1 min, after which the volume remains constant for the rest of the experiment. Therefore, V_bag during and after filling can be expressed by Equations (3) and (4):

$${\mathrm{V}}_{\mathrm{b}\mathrm{a}\mathrm{g}}^{\mathrm{n}}={\mathrm{V}}_{\mathrm{b}\mathrm{a}\mathrm{g}}^{\mathrm{n}-1}+{\displaystyle \frac{2}{60\times 1000}}\left({\mathrm{m}}^3\right)\hspace{1em}\mathrm{f}\mathrm{o}\mathrm{r} \mathrm{t}<1 \mathrm{m}\mathrm{i}\mathrm{n}$$

(3)

$${\mathrm{V}}_{\mathrm{b}\mathrm{a}\mathrm{g}}=\sum _{\mathrm{n}=1}^{60}{\mathrm{V}}_{\mathrm{b}\mathrm{a}\mathrm{g}}^{\mathrm{n}}\hspace{1em}\mathrm{f}\mathrm{o}\mathrm{r} \mathrm{t}>1 \mathrm{m}\mathrm{i}\mathrm{n}$$

(4)

The mass of water in the system can be represented by Equation (5):

$${\dot{\mathrm{m}}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{i}\mathrm{n}}(\mathrm{t})-{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{l}\mathrm{o}\mathrm{s}\mathrm{s}}\left(\mathrm{t}\right)={\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}\left(\mathrm{t}\right)-{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}\left(\mathrm{t}-1\right)$$

(5)

Isolating ${\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{l}\mathrm{o}\mathrm{s}\mathrm{s}}$, we get Equation (6):

$${\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{l}\mathrm{o}\mathrm{s}\mathrm{s}}\left(\mathrm{t}\right)={\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}\left(\mathrm{t}-1\right)-{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}\left(\mathrm{t}\right)+{\dot{\mathrm{m}}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{i}\mathrm{n}}(\mathrm{t})$$

(6)

We assume that the mass of water lost is given by Equation (7):

$${\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{l}\mathrm{o}\mathrm{s}\mathrm{s}}\left(\mathrm{t}\right)={\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{a}\mathrm{d}\mathrm{s} \mathrm{b}\mathrm{a}\mathrm{g}}(\mathrm{t})+{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{d}\mathrm{i}\mathrm{f}\mathrm{f} \mathrm{b}\mathrm{a}\mathrm{g}}(\mathrm{t})$$

(7)

where ${\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}$ is the mass of water in the bag at different times, ${\dot{\mathrm{m}}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{i}\mathrm{n}}$ is the mass flow added by bubbling at each time t (g), and ${\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{l}\mathrm{o}\mathrm{s}\mathrm{s}}$ is the total mass of water lost by both adsorption and diffusion (g).

The two independent experiment configurations result in a system of equations. By comparing the trend of humidity measured inside the bag in the two experiments, the contributions of water adsorbed and diffused can be easily isolated using the mass balance derived from Equations (6) and (7). Equation (8) refers to the first series of experiments using a simple polymeric bag (subscript 1), and Equation (9) refers to the second set of experiments (subscript 2), where the term due to the adsorption of the additional film is added.

$${\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{a}\mathrm{d}\mathrm{s} \mathrm{b}\mathrm{a}\mathrm{g}}{\left(\mathrm{t}\right)}_{\left(1\right)}+{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{d}\mathrm{i}\mathrm{f}\mathrm{f} \mathrm{b}\mathrm{a}\mathrm{g}}{\left(\mathrm{t}\right)}_{\left(1\right)}={\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}{\left(\mathrm{t}-1\right)}_{\left(1\right)}-{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}{\left(\mathrm{t}\right)}_{\left(1\right)}+{{\dot{\mathrm{m}}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{i}\mathrm{n}}\left(\mathrm{t}\right)}_{\left(1\right)}$$

(8)

$${{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{a}\mathrm{d}\mathrm{s} \mathrm{b}\mathrm{a}\mathrm{g}}\left(\mathrm{t}\right)}_{\left(2\right)}+{{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{d}\mathrm{i}\mathrm{f}\mathrm{f} \mathrm{b}\mathrm{a}\mathrm{g}}\left(\mathrm{t}\right)}_{\left(2\right)}+{{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{a}\mathrm{d}\mathrm{s} \mathrm{f}\mathrm{i}\mathrm{l}\mathrm{m}}\left(\mathrm{t}\right)}_{\left(2\right)}={{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}(\mathrm{t}-1)}_{\left(2\right)}-{{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}\left(\mathrm{t}\right)}_{\left(2\right)}+{{\dot{\mathrm{m}}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{i}\mathrm{n}}\left(\mathrm{t}\right)}_{\left(2\right)}$$

(9)

where ${\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{a}\mathrm{d}\mathrm{s} \mathrm{b}\mathrm{a}\mathrm{g}}$ is the mass of water absorbed by the bag (g), ${\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{d}\mathrm{i}\mathrm{f}\mathrm{f} \mathrm{b}\mathrm{a}\mathrm{g}}$ is the mass of water diffused by the bag (g) and ${\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{a}\mathrm{d}\mathrm{s} \mathrm{f}\mathrm{i}\mathrm{l}\mathrm{m}}$ is the mass adsorbed by the sheet of film inserted in the bag (g).

We assume that the contributions of water loss due to the bag’s wall are the same in both experiments, as the bags are identical in geometry and materials. By subtracting (4) and (5), we obtain Equation (10).

$${\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{a}\mathrm{d}\mathrm{s} \mathrm{b}\mathrm{a}\mathrm{g}}(\mathrm{t})={{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{a}\mathrm{d}\mathrm{s} \mathrm{f}\mathrm{i}\mathrm{l}\mathrm{m}}(\mathrm{t})}_{\left(2\right)}={\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}{\left(\mathrm{t}-1\right)}_{\left(2\right)}-{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}{\left(\mathrm{t}\right)}_{\left(2\right)}-({\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}{\left(\mathrm{t}-1\right)}_{\left(1\right)}-{\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}{\left(\mathrm{t}\right)}_{\left(1\right)})$$

(10)

As the internal bag surface is equal to the film surface, ${\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{a}\mathrm{d}\mathrm{s} \mathrm{f}\mathrm{i}\mathrm{l}\mathrm{m}}$ is equal to ${\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{a}\mathrm{d}\mathrm{s} \mathrm{b}\mathrm{a}\mathrm{g}}$. Therefore, ${\mathrm{m}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{d}\mathrm{i}\mathrm{f}\mathrm{f} \mathrm{b}\mathrm{a}\mathrm{g}}$ can be calculated by substituting into Equation (4), considering that the inlet water mass flow rate (${\dot{\mathrm{m}}}_{{\mathrm{H}}_2\mathrm{O}}^{\mathrm{i}\mathrm{n}}$) becomes zero after the filling is completed. This allows us to obtain the amount of water adsorbed and diffused for each material over time.

2.4.3. Phase 3: Effect of Dehumidification on VOCs

Urine Headspace Analysis

Phase 3 of the study was dedicated to evaluating the impact of the dehumidification process on the potential loss of VOCs together with water vapor. First, a comparative assessment of overall VOCs retention across the different drying materials was performed. To do this, the dried headspace samples from the same pooled urine (Class P) were obtained by the proposed drying system using each polymeric material under testing: Nalophan^TM bag, HDPE bag (8 μm and 11 μm), LDPE bag (12 μm), Bio-PS bag, and the Nafion^TM dryer (as a reference). Each sample was prepared following the protocol described in Section 2.4.1 (bubbling 20 mL of urine, drying to ~20% RH using the given method), and the dried headspace was then characterized using a PID (Tiger XT 10.6eV, Ion Science, Fowlmere, UK) and the e-Nose prototype. The PID provides an estimate of the total VOC concentration expressed as isobutylene-equivalent ppm, while the e-Nose provides a composite sensor response pattern. Each e-Nose measurement lasts 10 min: 3 min of reference air to establish the baseline, 4 min of urine headspace analysis, and 3 min of reference air to return to the baseline. By examining the PID readings and the e-Nose sensor response magnitudes, rough information about the total VOC concentration of the dried mixture was obtained. The results obtained from samples prepared with different materials were compared to identify the polymeric bags least prone to VOC loss during the dehumidification phase, with the goal of minimizing the loss of diagnostically relevant information in e-Nose-based clinical applications. In practice, Nalophan^TM (a widely used material for breath/urine sampling) was used as the baseline for “acceptable” performance, and the Nafion^TM dried sample was used as the reference for dehumidification. We looked for bag materials whose PID and e-Nose signals were closest to these references. This initial screening allowed us to narrow down the candidate materials to those that combined fast dehydration with preservation of VOC content similar to the references.

In the second part, the e-Nose was trained to discriminate between Class P (pooled urine) and Class A (spiked urine) samples dried using different drying methods (i.e., the best-performing sampling bag material identified in Phase 1 against Nafion^TM membrane dryer and Nalophan^TM bags, respectively). For each of these three sample preparation methods, we conducted multiple e-Nose measurements on both Class P and Class A samples. In total, three independent pooled urine batches (see Section 2.3.2) were used, each providing one Class P and one Class A sample. In total, eight repetitions with each sample (aliquot) type (i.e., three pooled raw samples and three spiked samples) and each material were performed. This provided a sufficient dataset to train and validate a preliminary classification model.

Data Processing and Statistical Analysis

Resistance time series were acquired with a LabVIEW^® 2025 Q3 interface, converted to CSV format, parsed, and processed with custom Python (v3.7.4) scripts. The changes in resistance were recorded and normalized using the fractional difference in resistance formula: ΔR/R₀, where R₀ is the baseline sensor resistance under reference air, and ΔR is the difference in resistance during exposure to the urine sample. Then, the time series were autoscaled to the unit variance, which refers to mean centering, and then divided by the standard deviation [47].

The autoscaled data was visually inspected using classical Principal Component Analysis (PCA) to evaluate the distribution of variance. The score plot (PC1xPC2) is shown in the Supplementary Materials (Figure S6). At this stage, no clear data separation between pooled (P) and spiked (A) urine samples was observed. Instead, separation due to the period of analysis (i.e., samples analyzed in November 2024 and February 2025) was observed. Therefore, the resulting matrix was subjected to Orthogonal Signal Correction (OSC) to eliminate systematic variation unrelated to the differences between Class P and Class A samples, such as temporal drift across data collected in different years (2024 and 2025). OSC is a pre-processing technique that removes variation in the predictor matrix (i.e., time series) that is orthogonal to the response variable (i.e., class label), thereby eliminating irrelevant information, such as temporal drift [48,49].

In this study, the optimal number of OSC components to subtract was determined based on the inter- and intra-cluster distance in the PC1 feature space. Specifically, the Euclidean distance (measure of the distance between two points in a multi-dimensional space, and it is defined by $\mathrm{D}\left(\mathrm{X},\mathrm{Y}\right)=\sqrt{{\sum }_{\mathrm{i}=1}^{\mathrm{T}}{\left|{\mathrm{x}}_{\mathrm{i}}-{\mathrm{y}}_{\mathrm{i}}\right|}^2}$, where D(X, Y) is Euclidean distance and ${\mathrm{x}}_{\mathrm{i}}$ and ${\mathrm{y}}_{\mathrm{i}}$ are the coordinate values in dimension i of points X and Y, T is the number of dimensions) between the centroids of Class P and Class A clusters, as well as the centroids of both year clusters (i.e., 2024 and 2025), was calculated for varying numbers of OSC components. The idea has been inspired by [50], which used the method for choosing the kernel parameters for the Support Vector Machine (SVM) model. The number of OSC components that maximize the distance between class centroids indicates more separated clusters. In contrast, the number of OSC components that minimize the distance between the centroids of the analysis period clusters indicates the removal of temporal drift. Additionally, to evaluate clustering quality, the Davies–Bouldin index (DBI) was computed. The DBI measures the average maximum similarity of each cluster. Its calculation involves both the inter-cluster distances denoted and the intra-cluster separation degree, as shown in Equation (11).

$$\mathrm{D}\mathrm{B}\mathrm{I}={\displaystyle \frac{1}{\mathrm{N}}}\sum _{\mathrm{i}=1}^{\mathrm{N}}{\max }_{\mathrm{i}\ne \mathrm{j}}\left({\displaystyle \frac{{\mathrm{s}}_{\mathrm{i}}+{\mathrm{s}}_{\mathrm{j}}}{{\mathrm{d}}_{\mathrm{i},\mathrm{j}}}}\right)$$

(11)

where N is the number of clusters, d is the Euclidean distance between the i-th cluster and j-th cluster, and s_i represents the average distance of all samples of the i-th cluster to the cluster centroid. This metric is positive, where a lower score generally indicates a better clustering [51].

Sample classification was performed using a linear SVM. SVM classifier optimization and performance assessment were based on double cross-validation [52]. The dataset for each sample preparation method was randomly divided into a training set (70%) for model tuning and a testing set (30%) for validation. Internal validation was conducted using five-fold cross-validation, while external validation was assessed using the test samples. The classification performance was measured in terms of the accuracy index, defined as the number of correct predictions divided by the total number of predictions.

3. Results

3.1. Phase 1: Dehumidification Performance

3.1.1. Effect of Dry Air Flow Rate on the Dehumidification Rate of Sampling Bags

The influence of dry air purge flow rate on the dehumidification rate of sampling bags was investigated. Figure 2A shows the humidity decay profiles after the initial 15 min filling period for a sampling bag made from LDPE (12 μm), exposed to different external dry air flow rates (0, 1, 5, and 10 L/min). The introduction of dry air purge externally to the bag significantly accelerated moisture removal compared to static conditions (0 L/min). Using a 1 L/min purge reduced the time to reach the desired humidity level from more than 2 h (no purge) to about 40 min. However, increasing the purge flow above 1 L/min did not further improve the dehumidification rate, indicating a plateau; above this flow, moisture removal was likely limited by water transport through the polymer governed by the material’s intrinsic permeability [44,53]. Based on these results, a flow rate of 1 L/min was selected to characterize the dehumidification time of all sampling bag materials in subsequent tests.

3.1.2. Dehumidification Rate of Sampling Bags Exposed to 1 L/min of Dry Air

Figure 2B shows the RH decay over time of humid gas samples stored inside different sampling bag materials exposed to a 1 L/min dry air flush. In most tested materials, the water content decreased from 70 to 80% to less than 20% within 1 h, except for LDPE (50 μm), which did not reach the target RH during the experiment, indicating reduced permeability due to its higher thickness. Among the materials, Bio-PS exhibited the fastest dehumidification (11 ± 3 min), while Nalophan^TM required slightly longer than the other materials (52 ± 3 min). Interestingly, despite identical filling procedures (i.e., 15 min bubbling distilled water to generate the initial humid air), the Bio-PS bags showed lower initial RH at the end of the filling step (i.e., t = 0) compared to the other polymers. This is likely a structure–transport consequence of polar ester groups of the Bio-PS that increases the affinity (solubility/sorption) for water, and its amorphous regions that provide free volume, a prerequisite for chain movement that drives gas diffusion [54]. Such polymer properties may explain the significant water loss already in the first minutes of filling. By contrast, non-polar polyethylenes (HDPE and LDPE) have very low water affinity, so even though they can have relatively fast diffusion due to their amorphous regions [2], their water vapor permeability remains low because permeability is dominated by the solubility term (P ≈ D·S) [55]. Nalophan (PET) combines polar ester groups with a rigid aromatic backbone, relatively constrained amorphous phases, and often significant crystallinity [3]; this typically yields much lower loss for larger penetrants (including many VOCs) and comparatively lower water vapor permeation than more flexible aliphatic polyesters (e.g., PBAT and PBS) present in the blend of Bio-PS. LDPE 50 μm, which reached the target RH extremely slowly (96 ± 8 min), compared with other polymers, was excluded from further experiments due to insufficient dehumidification performance. Additionally, to simplify subsequent testing while still capturing the effect of film thickness, the two intermediate HDPE thicknesses (9 and 10 μm) were excluded.

3.2. Phase 2: Investigation of Water Diffusion and Adsorption in Sampling Bags

Phase 2 of the study focused on investigating water diffusion and adsorption mechanisms during sample dehumidification to quantify their respective contributions. Following the observation of a rapid decrease in water vapor concentration within all bags after only a few minutes, the question arose as to whether this loss was primarily due to adsorption onto the bag walls or to diffusion through the polymeric material. To estimate the relative influence of these two phenomena on the observed reduction in humidity, a mass-balance model based on two complementary experiments was applied (Equation (10)).

Figure 3 and

Table 1. Mass of water (in grams) lost by polymeric sampling bags after 7 h of conditioning at 60 °C and 5% RH. The contributions from adsorption and diffusion were estimated using a mass-balance model.

	HDPE 8 μm	HDPE 11 μm	LDPE 12 μm	LDPE 50 μm	NalophanTM 20 μm	Bio-PS 15 μm
mass adsorbed (g)	0.0020 ± 0.0002	0.0020 ± 0.002	0.0003 ± 0.001	0.0002 ± 0.001	0.0020 ± 0.001	0.0016 ± 0.001
mass diffused (g)	0.2410 ± 0.0150	0.2369 ± 0.0144	0.2454 ± 0.010	0.2371 ± 0.009	0.2332 ± 0.011	0.2417 ± 0.009
mass lost (g)	0.2430 ± 0.0160	0.2389 ± 0.0139	0.2457 ± 0.010	0.2374 ± 0.009	0.2352 ± 0.011	0.2422 ± 0.012

show the estimated mass of water lost by diffusion and adsorption during conditioning under static conditions (no purge, 60 °C, 5% RH chamber). Overall, all materials lost about 0.24 g of humidity after 7 h of storage. Diffusion through the polymer emerged as the dominant mechanism driving the dehumidification process, with the adsorption losses being relatively small. These results are in accordance with literature findings [38,39,56].

Moreover, the results showed that rapid water adsorption on the bag’s wall occurs in the first minute of the experiment, when the bag is filling, whereas diffusion remains slow. Subsequently, adsorption levels decreased sharply as diffusion became dominant until equilibrium was achieved. Most of the materials presented a similar diffusion behavior, except for Bio-PS e and LDPE 50 μm, which showed noticeable deviations. For Bio-PS, the lower initial RH and steeper early-time mass loss are consistent with the strong water affinity noted earlier in Phase 1, whereas LDPE 50 μm, owing to its nature and larger thickness, showed a slower overall diffusion rate. The most dynamic phase of the dehumidification process occurred within the first hour of the experiment, when the humidity loss rate was highest. Material thickness significantly influenced diffusion rates, with thinner materials demonstrating faster moisture transfer due to shorter diffusion paths, in accordance with Fick’s law. This behavior reinforces the importance of considering both polymer chemistry and film thickness when designing polymeric barriers for moisture-sensitive applications.

3.3. Phase 3: Impact of Dehumidification on VOC Detection by e-Nose

3.3.1. Exploratory Analysis of Sensor Response

Figure 4 shows the PID readings and e-Nose sensors’ responses for pooled urine headspace samples after dehumidification with materials under test (Nalophan^TM, HDPE 8 μm, HDPE 11 μm, LDPE 12 μm, and Bio-PS bags) as well as the Nafion^TM dryer. Figure 4A compares the PID-measured total VOC concentration in the sample, expressed as isobutylene-equivalent ppm according to the PID calibration, before and after drying with each method, while Figure 4B shows the corresponding e-Nose sensor signal levels. The samples dried using the Nalophan^TM and Nafion^TM dryer retained the highest overall VOC levels—their PID readings after dehumidification were nearly as high as the original samples—indicating minimal loss of VOCs. Similarly, PID readings of samples dried using Bio-PS bags were nearly the same as the original undried sample; however, they presented lower PID readings before dehydration (i.e., right after the filling process), suggesting that VOCs were already lost during the 15 min of filling, and after dehydration, a few more were lost. This behavior is consistent with rapid sorption/solubility followed by diffusion of small oxygenated VOCs into the more polar and amorphous Bio-PS matrix [54]. The e-Nose sensors likewise showed strong response signals for Bio-PS and Nafion-dried samples, yielding e-Nose “fingerprints” comparable to the ones obtained using the Nalophan^TM bag, thus suggesting minimal loss or alteration of urine VOCs. Full sensor responses are reported in the Supplementary Materials (Figure S6).

In contrast, the samples prepared using polyethylene-based bags (HDPE and LDPE) exhibited significantly lower PID and sensor responses, suggesting that these materials allowed considerable diffusion losses of VOCs during sample storage [57,58,59]. This behavior can be attributed to the semi-crystalline structure of polyethylene [57], where amorphous regions provide pathways for gas diffusion [54].

It is important to underline that the PID provides an estimate of total VOC concentration and is thus useful for detecting potential analyte losses, but it does not give information on the qualitative composition of the mixture. In contrast, the e-Nose responses constitute an olfactory fingerprint that reflects the sample’s overall aroma profile. Therefore, the present Phase 3 data support conclusions about preservation of class-discriminative fingerprints, but not about the absolute retention of each analyte.

Based on the results obtained, the Nafion^TM dryer, Bio-PS bags, and Nalophan^TM were selected for subsequent evaluation of e-Nose class-discrimination performance.

3.3.2. Gas Clustering and Classification

The classification capability of the e-Nose to differentiate between pooled urine samples (Class P) and VOC-spiked pooled urine samples (Class A) was evaluated using the best-performing materials—Nafion^TM dryer, Nalophan^TM bag, and Bio-PS bag—identified from the preliminary tests.

In total, 48 samples (24 Class P and 24 Class A) from different batches were analyzed between 2024 (26 samples) and 2025 (22 samples) for each of the three materials. An exploratory analysis with PCA highlighted a strong temporal drift across analyses performed in different years (Supplementary Materials, Figure S7). Figure S7 shows that drift occurs primarily along the first principal component (PC1). The entire subset is completely shifted from the location of the first data subset. Therefore, in addition to variability and scattering due to local noise/drift, there is a long-term drift effect that displaces and changes the variance within the data set. Drift may be due to sensors’ aging and also to environmental changes that happened during the 4 months of measurements, since MOS sensors are strongly affected by temperature and humidity [60]. This shift, associated with sensor aging, makes the classification model developed on training data usable only within the same year.

OSC was applied to the training set (Class A: n = 17 and Class P: n = 17) to remove the temporal drift and maximize the difference between the two classes. OSC was applied to the time series dataset to enhance separation between Class P (pure) and Class A (spiked) samples by removing variance unrelated to the class labels. In addition to sample variability, temporal drift across data collected in different years (2024 and 2025) was also targeted by the OSC. The best number of OSC to be removed was determined by the inter-cluster distance of the training data, as shown in Figure 5. We employed the index DBI, calculated from class and year labels, and removed the resulting PC1 data for each OSC component. This metric provides a comprehensive assessment of clustering performance across various preprocessing methodologies, with lower DBI values indicating tighter intra-cluster aggregation and better clustering. For all three sample preparation methods, the removal of five OSC components maximized the distance between class centroids, while yielding a low DBI, and simultaneously minimized the separation between the year-based centroids, correcting drift associated with the time period in which the analyses were performed. Further removal of OSC components reduced class separation, suggesting that excessive correction could also eliminate useful class-discriminative information. The PCA obtained for the removal of one to five OSC components is reported in the Supplementary Materials (Figure S8).

After applying OSC (five components), we trained and tested SVM classifiers as described earlier.

Table 2. Summary of the SVM model performance for classifying urine headspace samples (pooled and spiked) following different sample preparation methods. Accuracy values are listed for the training phase (cross-validated) and the external test set (confidence limits 95% in brackets, calculated according to the binomial distribution) with parameter C = 0.01, using the (Rmin − R₀)/R₀ feature.

Sample Preparation Method	Number of OSC Components	Euclidean Distance Class	Davies–Bouldin Index	Accuracy Training CV (CI95%)	Accuracy Test (CI95%)
Bio-PS	0	2.235	1.918	0.743 (0.567, 0.875)	0.692 (0.385, 0.909)
	5	3.981	0.705	0.857 (0.699, 0.950)	0.769 (0.472, 0.936)
NalophanTM	0	1.726	2.403	0.739 (0.556, 0.876)	0.777 (0.460, 0.953)
	5	3.250	1.036	0.875 (0.719, 0.950)	0.833 (0.552, 0.953)
NafionTM	0	2.734	1.410	0.788 (0.691, 0.933)	0.833 (0.552, 0.953)
	5	4.046	0.702	0.808 (0.635, 0.923)	0.861 (0.559, 0.975)

summarizes the classification results achieved in double cross-validation and on the external test set for Nafion^TM dryer, Nalophan^TM bag, and Bio-PS bag. The Bio-PS bag method achieved a classification accuracy of 76% on the independent test set, the Nalophan™ bag yielded 83%, and the Nafion^TM dryer attained 86% accuracy in this validation. While the Nafion^TM dryer provided the highest accuracy, the Bio-PS and Nalophan^TM bag methods also maintained robust classification performance. At the same time, because each model was trained on a small dataset, these results should be interpreted as preliminary evidence that the main VOC fingerprint was preserved.

4. Discussion

This work aimed to develop a novel methodology for dehumidifying gas samples intended for e-Nose analysis of biological matrices. The proposed approach leverages low-cost, disposable polymeric materials purged with dry air to reduce sample humidity, thereby mitigating a major source of sensor interference while minimizing risks of cross-contamination. With this work, we aimed to establish a state-of-the-art method for studying new materials suitable for biological sample analysis. We structured our evaluation into three phases: dehumidification performance of the system (Phase 1), dehumidification mechanisms (Phase 2), and the impact on VOC detection (Phase 3). In Phase 1, various polymeric bag materials were evaluated for their dehumidification performance, considering the time required to reach target humidity (20% RH at 60 °C) levels under a dry air purge. This was achieved using a controlled drying system, in which the impact of 1, 5 and 10 L/min airflow rates was analyzed. The second phase focused on investigating the mechanisms of water diffusion and absorption in the selected polymeric materials. Experiments were conducted under controlled temperature and humidity conditions to characterize diffusion and adsorption phenomena. The final phase assessed how the different drying materials affected the measurable VOC fingerprint, using a PID and an e-Nose system to compare sensor responses and classification accuracy across the tested sample-preparation methods.

Among the tested sampling bag materials, Bio-PS exhibited the shortest dehumidification time (11 ± 3 min), which reduces the overall sample preparation duration and increases the efficiency of the analytical process. The lower initial RH observed for Bio-PS bag immediately after the 15 min of filling (about 40% RH) suggests a rapid short-time sorption of water into polar ester groups during filling, followed by fast film diffusion through its amorphous regions [54,61,62]. The same water affinity may also explain why part of the VOC loss occurred already during the filling phase, when small oxygenated compounds can be transported into the polymer before the active purge begins.

Nalophan^TM is another promising material for dehumidifying gas samples, requiring 52 ± 3 min to dry a saturated sample down to 20% RH. While slower than Bio-PS, the system still provided complete dehumidification within about 1 h, an improvement compared to the passive conditioning (which can take >2 h, depending on the bag volume). Importantly, Nalophan^TM retained the urine VOCs pattern sufficiently well to yield 83% classification accuracy, supporting its suitability for the conditioning of biological samples, as also suggested in previous studies on polymeric barriers for olfactory analysis [35].

In principle, the dehumidification time could be further reduced by increasing the surface-to-volume ratio (S/V) (i.e., by decreasing the bag volume) or by increasing the temperature. Assuming a diffusion-controlled process, the characteristic dehumidification time (${\mathrm{t}}_{\mathrm{d}\mathrm{r}\mathrm{y}}$) scales approximately with V/S and therefore for geometrically similar bags ${\mathrm{t}}_{\mathrm{d}\mathrm{r}\mathrm{y}}\propto {\mathrm{V}}^{1/3}$. Applying this scaling to Nalophan^TM (52 ± 3 min at 2 L), a target dehumidification time of 5 min would correspond to a bag volume of about 1.8 mL, i.e., an approximately 10x higher S/V would be required. This estimate indicates that reducing bag size alone is not a realistic strategy for achieving 5 min drying in e-Nose applications, because the required sample volume would be impractically small. Moreover, higher S/V is not necessarily favorable for compound preservation: in Nalophan bags, high bag surface-to-volume ratios were reported to decrease the compounds’ recovery [39].

Temperature is also expected to matter: higher temperature should accelerate water transport by increasing vapor pressure and polymer permeability. When the temperature increases, the free volume of polymers increases, which mainly affects the chain movement that will drive gas diffusion [54]. Temperature change will also alter the urine–headspace equilibrium and increase diffusion or release of compounds adsorbed to the inner walls of the sample bags [35,39,63]. In particular, higher temperatures may increase the loss of small oxygenated VOCs, such as acetone and 4-heptanone, through polyester materials, such as Nalophan^TM and Bio-PS [63]. Therefore, the bag material, the bag geometry, and the operating conditions should be jointly optimized to shorten dehumidification time while preserving the VOCs of interest.

In this study, the Nafion^TM dryer was examined as a benchmark. It proved very effective at continuously removing humidity from the samples and retaining VOCs, yielding the highest classification performance (86% accuracy in urine headspace discrimination). However, we observed that the dehumidification performance of the Nafion^TM dryer decreased over time with repeated use with urine samples. In fact, after a few cycles of use, the sample humidity at the membrane dryer outlet increased progressively with each cycle, even after extended dry air purging. It may be explained because the urine composition contains ammonia (NH₃), which can irreversibly react with the Nafion^TM membrane, limiting its proper regeneration/cleaning with dry air purge [64,65]. Nafion^TM can capture NH₃, converting it into ammonium ions (NH₄⁺), which bind strongly with its sulfonic acid network [65]. Over time, NH₃ and NH₄⁺ accumulate inside the Nafion^TM tubing, reducing its ability to transport water vapor. Additionally, the formation of NH₄⁺ may cause clogging risks to the downstream parts [66]. As a result, the membrane should be replaced to maintain high dehumidification performance. This need for replacement represents the main disadvantage of the system, given the high cost of the membrane [32]. Therefore, although Nafion-based systems showed excellent initial performance, their suitability for biomedical applications involving ammonia-rich samples, such as urine, may be limited. Further studies are needed to determine the number of effective operational cycles, the extent of performance recovery after extended dry air purging under the present operating conditions, and whether pretreatment steps to remove NH₃ could improve long-term use.

The Phase 2 experiments clarified the mechanism of water loss in the polymeric sampling bags. The derived mathematical model allowed the diffused and adsorbed fractions to be estimated directly from the paired RH-time experiments, without relying on literature diffusion coefficients or fitted permeability parameters. Under the present operating conditions, diffusion through the polymer’s walls was the dominant mechanism for humidity removal, whereas adsorption contributed mainly during the early stages of conditioning, in agreement with previous studies [39,63]. This is an important finding, as it indicates that the active purging strategy is effective because it enhances the humidity gradient across the bag wall and therefore directly promotes the dominant water-removal pathway. Additionally, the results showed that Bio-PS (15 $\mathsf{\mu }$m) exhibited the highest diffusion rate, whereas thinner materials (e.g., HDPE 8 and 11 $\mathsf{\mu }$m and LDPE 12 $\mathsf{\mu }$m) showed slower diffusion. This highlights the role of water–polymer interactions in addition to geometry. At the same time, while intrinsic polymer properties, such as crystallinity, crosslinking, and filler content, film thickness, S/V, and temperature influence water vapor transport, improper bag sealing can also lead to artificially high diffusion rates in some cases.

In Phase 3, the effect of the different drying materials on the measurable VOC fingerprint was assessed using a PID and an e-Nose system. In addition, we proposed using the inter-cluster distance metric to optimize the OSC drift-compensation step. This criterion made it possible to select the number of OSC components that maximized class separation while minimizing drift-related variance. Applying the chosen OSC parameters for each drying material resulted in a better cluster separation that corresponds to the true class differences (e.g., pooled vs. spiked urine samples), while removing temporal drift, consequently increasing the accuracy of the classifier. After OSC optimization, the best-performing polymeric materials yielded classification accuracies of 76% for Bio-PS and 83% for Nalophan^TM, compared with 86% for Nafion^TM. These results confirm that Nalophan^TM and Bio-PS sampling bags can preserve enough of the sample fingerprint to support discrimination between pooled and spiked urine while offering a cost-effective and disposable alternative to conventional dryers.

Finally, while the study involving spiked urine aimed to have a preliminary screening of potential polymeric materials that can be used for storage and dehumidification of urine headspace, it also demonstrates the potential of the proposed bag-based drying system. Future work should therefore optimize drying parameters, such as purge flow rate, bag geometry, and temperature together, test new polymeric materials, conduct long-term stability tests, and validate the most promising sampling configurations on a real diagnostic task, developing classification models for a real pathology (e.g., prostate cancer) with the goal of verifying that this sampling method does not affect the sensor performance and diagnostic capability negatively.

In conclusion, this work contributes to advancing sample preparation methodologies for e-Nose technology, offering a viable alternative to existing drying methods and with potential applications in medical diagnostics and other fields requiring precise VOC analysis. These results align with the ongoing advancements in e-Nose technologies, which are increasingly being recognized for their potential in non-invasive diagnostics and environmental monitoring [1].

5. Conclusions

This study presents a novel methodology for the dehumidification of gas samples intended for e-Nose analysis using actively purged polymeric sampling bags. This is a preliminary study aimed at an initial screening of various materials. As such, the tests were conducted using a “simplified” experimental procedure using pooled vs. VOC-spiked urine, without any real reference to a real-world application.

We showed that this approach can significantly reduce the time required to dehumidify humid gas samples compared with passive storage, decreasing sample preparation time from >2 h to <1 h, while preserving sufficient urine VOC fingerprint information for e-Nose discrimination. Specifically, exposing samples contained in HDPE, LDPE, Nalophan^TM and Bio-PS bags to a 1 L/min dry air purge enabled reaching the target humidity of 20% RH quickly, while polymeric materials such as Nalophan^TM and Bio-PS proved best overall balance between dehumidification time and minimal VOC loss. Compared with the Nafion^TM dryer, the bag-based approach preserved the sample VOC fingerprint important for the e-Nose classification performance, while avoiding the drawbacks of cost and cross-contamination associated with the multi-use of Nafion^TM dryers. Therefore, although Nafion^TM showed an excellent initial dehumidification performance, we advise against its repeated use for ammonia-rich matrices, such as urine, as ammonia in the sample may contaminate the membrane and progressively reduce its performance over time.

Analysis of the humidity profiles, supported by the derived mathematical model, indicated that water loss was governed predominantly by diffusion through the polymer wall, while adsorption effects were mainly limited to the initial stage of conditioning. These findings help clarify the mechanisms underlying the different storage behaviors of the tested materials and support the use of active purging as an effective strategy to enhance the dominant water-removal pathway.

Overall, the proposed bag-based methodology offers an affordable, disposable, and cross-contamination-free solution for mitigating humidity interference in e-Nose analysis of biological samples. Future research is necessary, considering a real-world diagnostic application and evaluating the effects of different sampling systems on diagnostic accuracy in real applications.