Real-time breath gas analysis of methane using a multipass cell-based near-infrared gas sensor

We demonstrated a near-infrared exhaled breath sensor for real-time methane measurements by using tunable diode laser absorption spectroscopy (TDLAS), which can enable the noninvasive diagnosis of intestinal tract problems. The core component of the near-infrared TDLAS sensor is a two-mirror-based multipass cell with nine-circle patterns. An optical path length of 23.4 m was achieved in a volume of 233.3 cm3, which effectively improved the detection sensitivity and shortened the gas exchange time. The minimum detection limit was 0.37 ppm by applying wavelength modulation spectroscopy, which was 12.4 times greater than that of direct absorption spectroscopy. In addition, combined with wavelength modulation spectroscopy, the two-mirror-based multipass cell enabled sub-second gas exchange time of 0.6 s. Methane breath experiments were conducted with six volunteers, and the real-time measurement results and concentrations at the end of exhalation were analyzed. This study demonstrates that the developed sensor has high sensitivity, high selectivity, and fast response for breath methane measurements and has promising potential for noninvasive, real-time, and point-of-care disease diagnosis in clinical applications.


Introduction
Breath gas analysis (BGA) is a rapidly developing and promising field for medical diagnostics and metabolic status monitoring that examines the molecular composition of exhaled breath gases [1][2][3].Hundreds of molecules are present in exhaled breath, which predominantly comprises 78% nitrogen, 15-18% oxygen, 4-6% carbon dioxide, 5% water vapor and 1% argon.In addition, hundreds of volatile organic compounds are present in concentrations of parts per million (ppm) or less [4][5][6].To date, some major volatile organic compounds have been considered characteristic biomarkers for specific diseases and metabolic disorders [7][8][9].Methane is the end-product of microbial fermentation in the gastrointestinal tract and is produced only by a nonbacterial subpopulation of the human microbiome [10].Elevated methane in breath may be related to abdominal bloating, constipation, vitamin B12 shortage, lactose intolerance, and other gastrointestinal conditions [11,12].
Real-time BGA imposes relatively high requirements on the instruments, which is due to the challenges of quantifying extremely small amounts of specific biomarkers and achieving sub-second gas exchange for well-designed systems [13].The classical techniques for BGA include mass spectrometry [14], electrical sensing [15,16] and gas chromatography [17].Mass spectrometry is sensitive to trace gases, yet it has difficulty identifying the target components in mixture gases, and the instrument cost is high [18].Electrochemical sensors are usually smaller in size and comparatively cost-effective.However, they have difficulty identifying certain molecules accurately and need frequent calibrations [19].Gas chromatography has high detection sensitivity at ppm levels or less.However, the extraction and purification processes are complicated and time-consuming, and that method is not suitable for real-time or online measurements [20].
Laser spectroscopy is effective for BGA due to its high selectivity and sensitivity, noninvasive nature, and fast response.Cavity-enhanced absorption techniques, such as cavity ring-down spectroscopy [21] and integrated cavity output spectroscopy [22], realize ultrahigh sensitivity by using a high-quality cavity .However, the high-quality cavity is usually composed of a pair of high-reflectivity (> 99.98%) mirrors, and the volume of the gas chambers is large.Consequently, it is difficult to develop an inexpensive miniaturized system.Tunable diode laser absorption spectroscopy (TDLAS) is also effective for trace gas detection in human breath.Recently, hollow waveguide-based mid-infrared (MIR) gas sensors were developed for BGA [23,24].Hollow waveguides require only a very small sample volume of several mL, yet the optical path length (OPL) is short.Hollow waveguide-based TDLAS sensors usually use a MIR laser to achieve target detection sensitivity.
Multipass cells (MPCs) are the core components in TDLAS, which improve the detection sensitivity by increasing the effective OPL.In 2017, Ghorbani et al. [25] and Ghorbani et al. [26] used MIR lasers and circular MPCs with OPL of 399 cm for the real-time analysis of carbon monoxide isotopes, CO and CO 2 in breath gas.However, lasers and detectors are expensive [27][28][29], and optical fiber technology is not yet mature in MIR regions.In 2019, Lou et al. applied an near-infrared (NIR) laser and a traditional Herriott cell to detect CO 2 in human breath [19].The Herriott cell had an OPL of 20 m, a base length of 32 cm and a volume of 420 cm 3 .Only one circle pattern formed on each mirror, and the utilization of mirror surfaces was low.In response to the demand for trace gas measurements with low gas consumption, there is a trend towards miniaturisation of MPCs, which reduce the volume of MPCs to less than 100 ml or even smaller [30][31][32][33].By calculating and adjusting the design parameters precisely, the newly developed MPC with a dense pattern can achieve a long OPL with a shorter base length and a smaller volume, making it suitable for developing NIR sensors with high sensitivity, cost effectiveness and fast response.
In this work, real-time measurements of CH 4 in exhaled breath using an NIR gas sensor based on a dense patterned MPC were demonstrated by using TDLAS.A 1653 nm distributed feedback laser was used to target the absorption line of CH 4 .The nine-circle patterned MPC had a simple structure of two coaxial spherical mirrors and achieved an effective OPL of 23.4 m in a volume of 233.3 cm 3 .A long OPL and small volume can effectively improve the detection sensitivity and reduce the gas exchange time, respectively.We performed direct absorption spectroscopy (DAS) and wavelength modulation spectroscopy (WMS) techniques to assess the linearity, limit of detection and long-term stability of the developed sensor.Additionally, the results obtained by the DAS and WMS methods were compared.The sequences of CH 4 expirograms of six volunteers were measured and analyzed by the WMS technique.The developed NIR sensor has the advantages of high selectivity and sensitivity, fast response time and inexpensive cost and was applied for the real-time monitoring of CH 4 during tidal breathing.

Basic measurement principles
The fundamental principle of laser absorption spectroscopy is the Lambert-Beer law.When a laser beam passes through a gas medium, the relationship between the transmitted laser ray intensity I t and incoming laser ray intensity I 0 is where α v = PLXS(T)φ represents the spectral absorbance.Here, P, L and X are the pressure, effective optical path length and concentration, of the measured gases and S(T) and φ are defined as the transition line strength and line shape function, respectively.Under weak absorption conditions, the transmitted laser ray intensity simplifies to TDLAS-based sensor systems use mainly tunable semiconductor lasers with narrowband wavelengths, which are effectively tuned by the injection current and temperature.The TDLAS technique includes two main methods: DAS and WMS.DAS has the properties of high stability and simple operation and has been widely applied in trace gas detection.WMS can further improve the detection sensitivity by significantly decreasing 1/f noise [34].

Spectral line selection
The high-resolution transmission molecular absorption database (HITRAN) is a worldwide standard for simulating molecular transmission and radiation, which covers a wide spectral region from microwave to ultraviolet [35].The HITRAN-based absorption spectra of 5 ppm CH 4 , 5% CO 2 and 5% H 2 O at approximately 6047 cm −1 are plotted in Fig. 1 under the conditions of 20 m OPL and ambient temperature and gas pressure.The concentration of main components of exhaled gas and HITRAN-based absorption characteristics near 6047 cm −1 at temperature of 296 K and pressure of 760 Torr with an optical path-length of 20 m, are shown in Table 1.Interfering absorption from N 2 , O 2 and Ar is not observed [35].The absorption intensity of 5% H 2 O and 5% CO 2 was 10 percent of the absorption intensity of 5 ppm methane.H 2 O is the one main interfering gas, and a dryer was used to effectively decrease the concentration of H 2 O under high humidity.CO 2 is the other main interfering gas, it is necessary to deduct the effect of carbon dioxide absorption.We used non-negative least squares algorithm to remove the effect.
According to Lambert-Beer law, when more than one gas is present in the optical path, Eq. ( 1) can be written as follows: where K measured and A measured are given by [k 1 , k 2 ] and [A ch4 , A co2 ] T , k 1 and k 2 are the ratio of actual CH 4 and CO 2 concentration to the standard gas concentration, which are non-negative values.A actual is absorption signal obtained from actual measurements.A ch4 and A co2 represent the absorption signals of CH 4 and CO 2 at the standard concentration, respectively, and the values of k 1 and k 2 can be obtained by non-negative least squares algorithm with better solution [36][37][38]: The above calculations help to separate the absorption signal and get the concentration of CH 4 from the mixed absorption signal, the concentration of CH 4 is k 1 multiplied by the standard concentration.Excluding the interference of H 2 O and CO 2 , the CH 4 absorption line is separated and strong at approximately 6047 cm −1 [39]; therefore, the CH 4 sensor has a high selectivity.
The R3 triplet of the 2ν 3 band at ∽6046.95 cm −1 was selected, which is the most commonly used in NIR methane detection [40][41][42].A distributed feedback laser (NLK1U5FAAA, NEL) emitting at 1653 nm was used as the laser source to target the selected line.The injection current was scanned between 50 mA and 90 mA, and the emission wavenumber was tuned in a narrow range of 6046.5∽6047.5 cm −1 to measure the methane concentration.

Design of the MPC with dense patterns
The traditional method for designing MPCs, such as the classical Herriott cell [43,44] and White cell [45] , is ABCD matrix analysis with the paraxial approximation.However, concave mirrors with high curvature provide relatively large spherical aberrations, and matrix-based ray tracing may cause significant errors after multiple reflections and with long OPLs [45].Consequently, obvious deviations are generated between the calculated spot patterns and the experimental results.Indeed, aberrations can be calculated accurately, and a nonparaxial method based on line-sphere equations was developed for simulating chief-ray tracing.The method was detailed in our previous work [45][46][47], and both the near-axis and off-axis rays were simulated exactly.Dense spot patterns were achieved on mirror surfaces, which significantly increased the utilization areas of the mirrors compared to the circular or elliptical pattern formed on the Herriott cell.
A MATLAB program was coded to simulate laser ray tracing and calculate the position of each spot.The 9-circle pattern was selected from our previous design results [46], and the photograph of the spot patterns is shown in Fig. 2. The diameter and curvature of spherical mirrors coated with a high-reflectivity dielectric film are 50.8mm and 100 mm.The incoming beam entered the two-mirror-based MPC with a distance of 118.4 mm from the entrance hole, was reflected 198 times and exited through the same hole.The MPC achieved an OPL of 23.4 m in a small volume of 233.3 cm 3 , which satisfied the desired sensitivity and fast gas exchange for CH 4 expiration detection.

Structure of the sensor
We developed a highly sensitive NIR sensor based on a nine-circle patterned MPC.The experimental setup is depicted in Fig. 2, which comprises three parts: a data acquisition and signal processing unit, an optical part and a gas handling system.The data acquisition and signal processing unit contains a computer and a DAQ card (National Instruments, NI USB 6363).The DAQ card was connected to a computer, and a LabVIEW-based program was coded for laser scanning, laser modulation and data acquisition.Additionally, the real-time WMS signal was processed by a lock-in amplifier coded in the LabVIEW program.
In the optical part, a 1653 nm distributed feedback laser was driven by a commercial laser current and temperature controller (ILX Lightwave, LDC-3724C).The laser current and temperature were set to 69.98 mA and 34.42 • C, respectively, and the R3 transition of the 2ν 3 band was targeted at 6046.95 cm −1 .The laser beam was coupled to a fiber collimator and entered the MPC through an entrance hole.Under the re-entrance condition, the laser left the cell through the same hole and was subsequently detected by an InGaAs photodetector (Thorlabs, PDA10CS-EC).
In the gas handling system, exhaled breath gas and ambient air were inhaled sequentially into the MPC through a breath tube, including a disposable antibacterial filter and a home-built buffer tube.A flow meter (CS230A, Sevenstar) was installed between the buffer tube and the MPC inlet to control the gas flow rate.The MPC outlet was connected to a pressure controller (640B, MKS) and a vacuum pump (SY-400, IRONMAN), and the pressure was kept at ambient pressure.

Direct absorption spectroscopy measurement
In the DAS experiment, a sawtooth signal with a 1 V amplitude and 10 Hz frequency was utilized to sweep the laser frequency.The measured signal was acquired by a DAQ card and was processed by a LabVIEW program.First, pure N 2 was flushed into the MPC for approximately forty minutes to obtain a baseline signal.Subsequently, a standard gas of 50 ppm CH 4 was filled into the MPC.The measured signal of CH 4 and the baseline signal are shown in Fig. 3(a), and the results were averaged over two scans.The experimental absorption with the Voigt line shape profile is plotted in Fig. 3(b).The absorption in the time domain was transformed to the frequency domain by using a solid etalon.The peak value of the absorption signal was 0.0430, and the standard deviation of the signal in the nonabsorbance range was 0.0013 (3 σ=0.0039).Hence, the signal-to-noise ratio (SNR) was 10.8, and a minimum detection limit (MDL) of 4.6 ppm was obtained.The dynamic response is an important criterion to be assessed for online and real-time monitoring.Three steps were conducted to measure the gas exchange time.The MPC was first with high-purity N 2 , then filled with 50 ppm CH , and finally filled with N 2 again.The dynamic response of the sensor was tested at a flow rate of 30 standard liters per minute (slm).The volume of the compact MPC was approximately 233.3 cm 3 , which could effectively shorten the gas exchange time.As plotted in Fig. 4, the rise time (10%-90%) and the fall time (90%-10%) were both 0.6 s.

Wavelength modulation spectroscopy measurement
The WMS technique was performed to improve the sensitivity of the sensor.The distributed feedback laser wavelength was scanned by a 10 Hz, 1 V sawtooth waveform and was modulated by a 20 kHz, 0.65 V sinusoidal wave signal.Using the LabVIEW program, the second harmonic was extracted from the measured absorption signal.The second harmonic signal profile of 10 ppm CH 4 at ambient temperature and pressure is shown in Fig. 5, and the signal was averaged over two scans.The peak value of the absorption signal was 1.784 mV, and the standard deviation of the signal in the nonabsorbance range was 22.15 µV (3 σ=66.45 µV).The SNR and the MDL were 26.9 and 0.37 ppm, respectively, which were 12.4 times better than those in the DAS experiments.The linearity of the gas sensor was estimated by measuring a mixture of CO 2 at 5% concentration and CH 4 over a gradient concentration range of 5ppm to 20ppm (5 ppm, 7.5 ppm, 10 ppm, 15 ppm and 20 ppm respectively).The measured concentrations are shown in Fig. 6(a), and each concentration level was measured for 550 s.The data for each measured CH 4 concentration level were averaged, plotted and fitted versus the standard gas concentrations, as shown in Fig. 6(b).Under conditions of interference with carbon dioxide gas absorption, an R-square value of >0.999 was obtained, which demonstrated the CH 4 sensor combined with the non-negative least squares algorithm has excellent linearity.
The detection precision and long-term stability of the developed sensor were assessed by Allan-Werle deviation analysis [48].We measured the CH 4 concentrations in the standard gase for 400 s, and the results are plotted in Fig. 7(a).The standard gase with 5 ppm CH 4 and 5% CO 2 concentration level was prepared by diluting the high-purity nitrogen with the certified 50 ppm CH 4 and certified pure CO 2 , respectively.The Allan-Werle deviation is demonstrated in Fig. 7(a), and the detection precision was 11.4 ppb at an integration time of 54.2 s.Until the integration time increases to 54.2 s, the system noise is mainly white noise, and the Allan deviation gradually decreases as the integration time increases, and when the optimum integration time is reached, further averaging will not help to improve the detection precision.When the integration time is longer than 54.2 s, the Allan-Werle deviation will start to deteriorate as a result of drift effects and changes in the spectral background structure, which is consistent with the trend of the Allan-Werle deviation in other studies [49][50][51].The detection precision was comparable to those of previous works [33,52,53].The histogram of methane concentrations is illustrated in Fig. 7(b), and the concentration distribution followed a Gaussian distribution with 1 σ and half-height half-width valued at 76.6 ppb and 90.3 ppb, respectively.The high detection precision demonstrated the high performance of the designed sensor based on the nine-circle patterned MPC.

Real-time measurement of the breathing cycle
Before the experiments, we notified the volunteers that they should eat a light diet, rinse their mouths, and conduct the tests under resting conditions to changes in the composition of exhaled gas due to other exogenous intake such as eating or other unrelated factors.Real-time detection of CH 4 exhalation and inhalation processes was performed using a volunteer.The volunteer sat relaxed and implemented normal breathing and deep breathing.The pressure and gas flow rate in the MPC were kept at ambient pressure and 30 slm, respectively.Consecutive profiles for CH 4 concentrations were obtained by using the WMS technique, and no data smoothing was performed in this work.The sequence diagrams of the CH 4 expirograms for the subject is shown in Fig. 8, and the measured concentrations of end-tidal exhaled CH 4 were around 10.0 ppm in four normal breathing and up to 13 ppm in deep breathing.During the inhalation period, ambient air was immediately pumped and flushed into the MPC through the breath tube.In the exhalation stage, the participant showed a rapid increase in exhaled CH 4 concentration.Moreover, the sharp decrease at the end of exhalation of the volunteer demonstrated the fast response time of the developed sensor.
In addition, we performed real-time online measurement of CH 4 in the breathing of another five subjects, and plotted the concentrations of end-tidal exhaled CH 4 for all subjects during deep breathing in Fig. 10.As shown in the figure, the respiratory CH 4 concentration levels of four subjects were lower than the CH 4 concentration in the air, one subject's respiratory CH 4 concentration was equal to the level in the air, and the respiratory CH 4 concentration of another subject was significantly higher than the CH 4 concentration in the air.On the basis of the exhaled breath CH 4 concentration [11,54], the volunteer was considered a high CH 4 emitter as the CH 4 concentration exceeded 5 ppm, while the remaining volunteers were classified as low emitters.A high level of exhaled CH 4 may be related to gastrointestinal problems [10][11][12][13], such as vitamin B12 deficiency and lactose intolerance.The developed sensor has outstanding performance with high detection sensitivity, a small volume, and fast response, which makes it well suited for the real-time clinical diagnosis and on-site monitoring of BGA.

Conclusion
A NIR exhaled breath sensor based on the dense-patterned MPC for real-time CH 4 monitoring was developed using the TDLAS technique.As the core component of the TDLAS sensor, the MPC formed nine-circle patterns on two spherical mirrors and achieved a long OPL of 23.4 m and a small volume of 233.3 cm 3 .Consequently, the sensor has the advantages of high selectivity and sensitivity, fast response, light weight and cost-effectiveness and is suitable for noninvasive real-time breath monitoring and clinical diagnosis.The minimum detection limit using the WMS method was 12.4 times better than that using the DAS method.Using the WMS method, the sub-second gas exchange time was 0.6 s and the detection precision of CH 4 was 11.4 ppb.Sequences of CH 4 expirograms were measured from six volunteers by using the WMS technique, and real-time monitoring of CH 4 was achieved during tidal breathing.We identified and quantified breath CH 4 with sub-second measurement time, which can enable the noninvasive diagnosis of intestinal problems.With the development of the NIR MPC-based sensor, clinical disease diagnosis and respiratory monitoring for noninvasive health screening tests may be possible in the near future.

Fig. 3 .
Fig. 3. DAS measurement of 50 ppm CH 4 in the MPC.(a) Raw data traces with and without CH 4 absorption and (b) measured absorbance along with the Voigt fitting.

Fig. 6 .
Fig. 6.(a) Measured CH 4 concentrations at different concentration levels and (b) linear relationship between the measured concentrations and the standard gas concentrations.

Fig. 7 .
Fig. 7. (a) Measured CH 4 concentrations in ambient air (top) and Allan-Werle deviation stability of the sensor (bottom) and (b) histogram of CH 4 concentrations.

Fig. 8 .
Fig. 8. Sequence diagrams of the CH 4 expirograms for a volunteer by using the WMS technique.

Fig. 9 .
Fig. 9. CH 4 expirogram during a breath cycle of the first subject and three different exhalation phases.
c e n t r a t i o n s ( p p m ) V o l u n t e e r N o .

Table 1 . The concentration and HITRAN-based absorption characteristics of exhaled gases near 6047 cm −
where n represents the number of types of multi-component gases and k n represents the scaling factor coefficient of the concentration of the nth gas compared to the standard concentration.If both methane and carbon dioxide satisfy the weak absorbance condition, the absorption signals of the two gases and the concentration are positively and proportionally correlated, such as their 2f signal amplitude.The real measured absorption signal can be regarded as a linear superposition of the standard absorption signals of CO 2 and CH 4 , as shown in Eq. (4): 1Gas species Concentration Line Intensity (cm −1 /(molec • cm −2 ))