Compact laser spectroscopic sensor head prototype for time-resolved breath oxygen monitoring

A small and lightweight optical sensor head prototype with a disposable airway adapter for continuous mainstream monitoring of oxygen at high sampling rate is designed and tested on an optical benchtop. In terms of its size and functionality, the sensor head design is similar to current capnography systems from leading medical equipment manufacturers, and it has been designed within constraints of potential applications in direct breath oxygen monitoring that require direct interaction with the gas inside a breathing tube. The measurement precision of 0.1% O2 with a 10 ms integration time are well within the performance required for breath O2 monitoring applications.


Introduction
Monitoring of the rate of oxygen consumption in critical care patients provides unique insight into metabolic function and may therefore be of use in a variety of clinical settings. Hospital rooms are fitted with many vital sign monitors: heart rate, blood oxygenation, blood pressure, body temperature, etc. However, the rate of oxygen consumption while breathing is not measured routinely. Assessing patients' metabolic function in real time is useful in several clinical dilemmas: neo-natal complications, sepsis identification, hypoxia detection, and nonresponsive patient monitoring [1][2][3].
Real time breath measurement must be sufficiently accurate and precise in order to clearly resolve the trends within each breath that are indicative of various ailments [4][5][6][7]. Devices used in medical settings must also be user-friendly, compact, and robust [8], with minimal maintenance requirements. Devices must also utilize a sterile, replaceable adapter [9] that could either be inexpensive and disposable, or easy to sterilize between use on different patients.
However, time-resolved and accurate measurement of O 2 in each human breath cycle with a desired precision of <1% and a sampling rate of 100 Hz [9] places high demands on the oxygen sensor itself. So far, commercially available techniques for oxygen monitoring in breath are bulky [8], limited in accuracy [10], or have slow response times [6,11].
Many of these issues can be effectively addressed by the recent advances of laser absorption spectroscopy (LAS) [12][13][14][15], whose strengths are high temporal resolution, selectivity, and sensitivity, all provided in a non-invasive manner. However, the small absorption cross-sections of O 2 transitions often need to be compensated with long absorption pathlengths or other optical enhancement techniques [8] to reach the desired minimum detection limits, which often lead to bulky sensors with large sampling volumes. This renders them less suited for routine mainstream breath monitoring in a clinical setting. Cavity enhanced absorption spectroscopy significantly increases the pathlength while maintaining a compact design, but requires many reflections between high-finesse mirrors [14], which is not compatible with intracavity windows required by replaceable breathing tube adapters. Photoacoustic spectroscopy (PAS) or quartz enhanced PAS [13] are promising alternatives that could potentially be further miniaturized to achieve high sensitivity while maintaining a small form factor; however, due to O 2 having poor relaxation dynamics strongly dependent on other collision partners (e.g. H 2 O vapor) as well as due to the need of direct interaction with the sample, PAS techniques are not ideal for achieving reliable sensor calibration and sterile operation. Many other forms of LAS enhancement such as optical feedback or the application of frequency combs as spectroscopic sources can further improve the sensitivity, but they also introduce complexity and cost that does not favor a small lightweight sensor for breath analysis [8].
We have iterated through several designs that could potentially operate within constraints of a dedicated mainstream breath monitoring device, and the results of this development process are presented below.
The ideal form factor for the oxygen sensor has been driven by the capnography instruments that are routinely used for breath monitoring via airflow rate and CO 2 concentration measurements. Broad capnography use has been driven mainly by increased adoption by physicians, especially in pediatrics and geriatrics, where early detection of respiratory failure is vital and the scale of this adoption is clearly visible in the global capnography market that is expected to see an annual growth of 8% [16]. Providing an oxygen sensing platform as a complement to the established CO 2 sensors will further expand the routine breath analysis applications by enabling collection of additional metabolic function data [2].
The general view of clinicians has been that laser technology for breath analysis is in its infancy and that devices have performed poorly in practice [10]. To break away from this reputation, our oxygen breath sensor head prototype has been designed for simplicity and size first, and then optimized for precision while under target test conditions. In this paper, we present the design of an oxygen sensor head that meets the stringent requirements of mainstream breath analysis.

Sensor design
The target specifications necessary to be useful for clinical use are developed based on [2, 8-11, 17, 18] and by assuming a form factor of a typical capnograph sensor used for mainstream breath monitoring. The summary of the specifications is shown in table 1. These metrics provided the bounds for the following system design process.
The minimum sensitivity for O 2 breath analysis (1% as indicated in table 1) is not especially difficult to achieve if there are no size and functionality constraints, but the small form-factor and short averaging time (10 ms) present a challenge that has been difficult to accomplish with available sensor technologies. The best feasibility studies of a spectroscopic oxygen sensor in a clinical setting to date was performed by Ciaffoni et al [15] using cavity-enhanced spectroscopy method. Their prototype sensor head was still relatively large (1.9 kg and 17 × 14 × 14 cm) as compared to typical capnography sensors. More importantly however, a replaceable adapter providing a sterile separation from patient to patient has not yet been demonstrated using laser spectroscopy for oxygen measurement in breath. Therefore, in this work to achieve the desired sensor requirements, we have considered several spectroscopic techniques to achieve the performance, functionality and form-factor goals. These techniques include: Faraday rotation spectroscopy (FRS), wavelength modulation spectroscopy (WMS), cavity enhancement (CE) techniques, and multipass cell arrangements. Based on our preliminary experiments and our previous work [19][20][21], we concluded that FRS, despite offering several important advantages such as improved detection limits and insensitivity to common diamagnetic interferences in breath, would not be well suited for compact mainstream breath analysis due to an inability to maintain the polarization state after propagation through a windowed airway adapter required for the sensor's completely sterile operation. Additional optical interfaces that could cause reflections and scattering are detrimental to polarization quality and thus deteriorate system sensing performance, causing FRS to require mutipass, or optical cavity enhancement. Therefore, WMS LAS approach with a miniature multipass Herriott cell equipped with a windowed airway adapter was selected as the most viable and robust sensor head design to meet the requirements listed in table 1.
The oxygen A-band is used for spectroscopic sensing. The strongest oxygen Q P 5 (4) transition accessible within a tuning range of a compact distributed feedback (DFB) laser (Nanoplus, 2583/14-18) was targeted as shown in figure 1. A miniature multipass cell was constructed using two spherical mirrors, with a radius of curvature of 1 m and a 12.5 mm diameter and 3 mm through-hole, that are spaced ∼3.45 cm apart. The multipass cell was configured to have 21 passes, which resulted in an effective optical path of 70 cm.
The oxygen sensing was implemented using WMS technique applied to spectroscopic signal detection. The modulation depth and lock-in amplifier (LIA)  settings were optimized to maximize the 2 nd harmonic (2f) WMS output signal, resulting in a modulation depth of 2.2 × ∆ν FWHM and a LIA bandwidth of 500 Hz, and the laser frequency was stabilized using the 3 rd harmonic (3f) WMS zero crossing to assure a line-locked operation of the sensor. First the system was tested with a single 200 µm thick BK7 window added between the multipass cell mirrors for evaluation of its impact on the sensor performance. The single uncoated window reduced the light on the detector by~4 times. To mitigate the impact of intracell windows and to improve overall optical power budget, two 100 µm thick, anti-reflection (AR) coated windows (650-1050 nm) were incorporated in the setup, and despite the four additional optical interfaces, they provided only 3.75 reduction in optical power with the same reduction in O 2 sensing precision. The AR-coated windows met the requirements for the sensor head design with removable and disposable breathing tube adapter as shown in figure 2, while providing the compact form factor and the short-term precision within the required limits in table 1.  The removable adapter is held in place using magnetic attachment and is located in the ∼3.45 cm space between the multipass cell mirrors. The adapter windows are separated by ∼3 cm, which limits the effective optical interaction path with the breath to ∼0.6 m. The volume of the disposable adapter is ∼8 cm 3 . All system elements can be further integrated into a monolithic sensor chassis with dimensions <8 × 4 × 4 cm, and model renderings of such a design are schematically shown in figures 2(b) and (c). WMS signals while performing a laser scan over the Q P 5 (4) transition are shown in figure 3.
Mainstream breath measurements are expected to cause significant optical power fluctuations due to potential window condensation or aerosol scattering, therefore the 2f WMS signal is normalized to the 1st harmonic (1f ) for calibration-free measurement [22]. Note that the 2f /1f signal is multiplied by a constant factor of 101.9 obtained via system calibration to convert to O 2 concentration units. We have tested the efficiency of this 2f /1f normalization by inserting an uncoated 2 mm thick BK7 window in-between the laser and the disposable adapter module and by observing the O 2 signal shown in figure 4. For comparison, the 2f WMS signal without 1f normalization shows significant deterioration of accuracy when the window is inserted. The signal is reduced by a factor of ∼2 and the observed noise is increased due to the window being handheld, yet the 1f normalization maintains its accuracy and the 3f lock to the transition center. In addition, the light is completely blocked using an aluminum block to demonstrate the locking scheme's ability to quickly regain the lock after a complete loss of signal.

Final sensor prototype configuration
A block diagram of the system configuration is schematically shown in figure 5.
Based on the tests performed, the final configuration of the sensor head has been based on the 763 nm Nanoplus DFB diode laser collimated with an aspheric AR coated (650-1050 nm) lens. The laser frequency is modulated at 3.37 kHz with an optimized modulation depth of 7 GHz (∼2.2 ν m /ν FWHM ). The collimated beam is directed into the multi-pass cell through the hole in the first mirror. The multipass cell chassis is custom designed to provide magnetic attachments and self-positioning for the disposable adapter assuring repeatable alignment. A heating element is placed in contact with the O 2 sensor chassis to prevent condensation on the windows during measurement, which is a similar approach as in the CAPNOSTAT 5 (commercial laser-based sensor for CO 2 measurement in breath) that uses internal heating elements to prevent condensation. The disposable adapter has standardized inlet and outlet outer diameter sizes for breath airway adapters (15 mm and 22 mm) and two 100 µm-thick AR coated (650-1050 nm) windows placed on the ends of the adapter (∼3 cm apart) perpendicular to the airway that allows the laser beam to pass through. After 21 passes through the airway adapter, the beam exits through a hole in the second multipass cell mirror and is measured using a photodetector (Thorlabs SM052A) equipped with an aspheric focusing lens (Thorlabs AL1225M-B). The photodetector signal is amplified using a transimpedance amplifier (FEMTO DHPCA-100), whose output is demodulated using a LIA (Zurich MFLI). The LIA provides access to multiple harmonics of the demodulated signal and the 3rd harmonic is used to lock the laser to the transition and the 2nd and 1st harmonics are used to determine the O 2 concentration using the WMS 2f /1f normalization methodology. An Allan-Werle analysis of the WMS 2f /1f retrieved concentration of O 2 at 20.9% over 500 s is supplied in figure 6. This system now achieves a short-term sensitivity down to ∼0.04% O 2 at 10 ms averaging time (dark red line collected for real-time update rate of 418 Hz).
Due to hardware data storage limitations, in order to investigate long-term drift of the test-bench the update rate was reduced to 52 Hz (while LIA bandwidth was kept unchanged), which allowed for collection of longer time acquisitions for Allan-Werle   deviation analysis (shown figure 6(a) as a light blue line). As expected, due to ∼8x under-sampling, the short-term precision has been reduced by ∼2.8x, but the curve also clearly reveals a sensor drift for averaging times beyond 1s. To better show the longterm drift, modified 2-point deviation analysis that provides a direct measurement of the system accuracy as a function of time after calibration [20] is shown in figure 6(b). The observed drift has been attributed to parasitic etalon fringes in the system that will be mitigated in the future by using wedged windows. However, the fringe drift does not cause the sensor accuracy to exceed 0.2% (shown in dotted line), which is well-within specifications in table 1.  The accuracy of the sensor was further evaluated over the full dynamic range (0%-100%) using the calibration curve shown in figure 7. A mass flow controller based gas dilution system (Environics 4040) is used to change the oxygen content of the gas as it flows through the sensor and the sensor voltage normalized to 100% oxygen is plotted on the y-axis. The WMS signal is linear over the entire dynamic range and the standard deviation of the error is ∼0.9%. These measurements show that the system's accuracy remains below the target of 1% for both long time periods as well as over the full dynamic range of the sensor.

Mainstream O 2 sensing test
We have performed several expository measurements to demonstrate the practical performance of this mainstream sensor head prototype.
First, we evaluated the efficiency of 2f /1f normalization by testing the system in a breath measurement purposely affected by extensive condensation in the system. For this particular test the heating elements designed to prevent condensation were not installed. Figure 8 shows breath measurements with the O 2 trace acquired using conventional 2f WMS clearly shows abnormalities in the breathing pattern in comparison to the 2f /1f normalized WMS time sequence data. Therefore, a final sensor head design employing both the heating elements to reduce condensation and 2f /1f normalization to further improve robustness to non-spectroscopic signal fluctuations, will ultimately provide the most reliable sensor operation.
In the second test, simulated breathing patterns were tested. To ensure better accuracy of mixing ratios, the simulated breathing patterns were generated using the Environics 4040 gas diluter. Fast shallow breaths in figure 9(a) are generated by the gas diluter and are measured with the sensor platform. In figure 9(b), the gas mixture is again used to simulate breathing where the concentration of oxygen is varied between 21% and 14% at time scale consistent with a breathing pattern, but on the 4th and 5th simulated 'breath,' the lower concentration is increased from the initial 14% to 16%.
The tests clearly confirmed that the system is able to resolve the O 2 differences at the sub-percent Figure 10. A prototype O2 sensor-head presented in conjunction with the CAPNOSTAT5 and a flow-meter (a). Real-time measurement of CO2 concentration using the CAPNOSTAT5 (b), mouth air pressure measured with the Respironics differential pressure flow sensor (c), and O2 concentration using the developed O2 sensor-head prototype (d) were all measured simultaneously while a healthy volunteer breathed through the entire assembly tested on an optical benchtop. level that is of interest for the relation to metabolic function.
Lastly, the sensor platform was run simultaneously with the CAPNOSTAT 5 CO 2 with an external mouth pressure meter. The O 2 sensor was placed in line with the CO 2 sensor and pressure meter as shown in figure 10(a). Then a volunteer breathed into the mask for several breath cycles as shown in figures 10(b)-(d). In this example data set the increase in CO 2 measured with a commercial capnograph (CAPNOSTAT 5) is consistent with increasing O 2 consumption as indicated by the O 2 trace measured with the developed sensor.

Summary
In this paper, we have presented a design and tests of a compact, LAS-based oxygen sensor head that meets all the requirements of a mainstream breath oxygen analyzer in a clinical setting. Initially, we assumed that significant improvements can be achieved by incorporating Faraday rotation and CE techniques to breath O 2 analysis, but after careful considerations of technological solutions required to provide an exchangeable airway adapter, implementation of this technology proved to be challenging in practice. In the process of system design and optimization, a miniature multipass cell enhanced WMS LAS sensor with 2f /1f normalization has proven to provide the necessary sensitivity and accuracy while enabling the compact form factor and required configuration that incorporates an exchangeable airway adapter. The designed disposable adapter, shown in figure 2 is of importance for medical applications that require sterile instrumentation. The chamber that the breath flows through must be inexpensive (disposable), as well as easily and robustly replaced on a regular basis by non-technical personnel. Therefore, our design achieves the target metrics in table 1 regarding sensitivity requirements while implementing an easily replaceable, disposable airway adapter that functions as the sample cell.
The sensor head prototype is small and lightweight and enables continuous mainstream monitoring of oxygen at high sampling rates. The prototype sensor head weighs less than 200 g, and has a warmup time less than 60 s, and can be fully integrated into <8 × 4 × 4 cm sensor head depicted in figure 2(b). The system uses a compact multipass cell to achieve ∼0.6 m effective optical path within a short physical dimension of ∼3 cm, and it is directly probing the exhaled air in mainstream configuration. The sensor can be easily integrated with existing capnography systems from leading medical equipment manufacturers. The achieved performance metrics are a measurement accuracy and precision below ∼0.2% O 2 with a time resolution of 10 ms (100 Hz sampling rate), which is adequate to fully resolve the oxygen concentration within a single breath cycle. The variance in measurement accuracy both over extended measurement times as well as over the full concentration dynamic range is less than 1%.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).