3.1.

Sol-Gel-Based Sensor Preparation

Oxygen-sensitive sol-gels were prepared by acid-catalyzed hydrolysis and condensation of the precursor propyltriethoxysilane (PTEOS). In a typical preparation, PTEOS was combined with absolute ethanol, followed by drop-wise addition of aqueous hydrochloric acid at pH 1. The mixture was then stirred for $20\min$ . An ethanolic solution of $\left[\mathrm{Ru}\right(\mathrm{dpp}{)}_3{]}^{2+}$ was subsequently added to the sol to give a final dye concentration of $2.5\mathrm{g}∕\mathrm{l}$ with respect to the total volume of the solution. The final mixture was magnetically stirred under ambient conditions and aged for several weeks before use. All chemicals were purchased from Sigma-Aldrich, Ireland, with the exception of the $\left[\mathrm{Ru}\right(\mathrm{dpp}{)}_3{]}^{2+}$ complex, which was synthesized at the National Centre for Sensor Research, Dublin.

3.2.

Optical Sensor Module

A schematic of the optical configuration employed in the breath analyzer is shown in Fig. 1 . The central element of the system is an injection-molded, polycarbonate sensor chip that was specifically designed and fabricated in-house for the efficient collection of luminescence from a thin sensor film. The chip consists of a frustrated cone structure that serves to redirect the anisotropic luminescence emission onto a photodetector positioned beneath the chip. Such a design can yield a significant $(\ge 25\text{-}\mathrm{fold})$ increase in the intensity of the collected luminescence compared with a conventional, planar chip.¹⁰ This has considerable implications for the design of low-cost, efficient sensors, and such technology is ideal for this application, combining a high degree of sensitivity with a disposable sensor element.

Fig. 1

Optical configuration of the sensor module. Inset shows the principle of operation of the sensor chip.

The portable, low-cost nature of the system is further enhanced through the use of a 5-mm-diam blue LED light source ( ${\lambda }_{\max }=450\mathrm{nm}$ , Roithner Lasertechnik, Austria) and a 8-mm Si PIN photodiode detector (S1223-01, Hamamatsu Photonics, United Kingdom). The LED provides direct excitation of a sensor spot that is deposited onto the upper surface of the frustrated cone, which directs the oxygen-sensitive, luminescence emission from the spot onto the active area of the photodiode (photodiode 1). A blue additive dichroic filter (Edmund Optics, United Kingdom) was placed in front of the LED, and a long-pass emission filter with a cut-on wavelength of $550\mathrm{nm}$ (Thorlabs, Newton, New Jersey) was placed between the sensor chip and photodiode to provide spectral discrimination of the excitation and emission signals. Background light levels at the detector were further reduced through the use of an optical aperture, which ensured that only the coated upper surface of the frustrated cone was exposed to the excitation radiation. The LED output was sinusoidally modulated at a frequency of $20\mathrm{kHz}$ , and the resultant, modulated luminescence emission was detected at the photodiode. Dedicated phase fluorometric detection electronics that were designed in-house were used to detect the phase shift between the excitation and emission signals, which is related to the lifetime of the luminescent complex, as shown in Eq. 2

3.3.

Deposition of Sensor Spots

The geometry of the substrate necessitates the use of a patterning technology that facilitates the deposition of discrete spots of material. Consequently, printing techniques are an ideal choice for sensor film deposition. Two techniques were investigated poly(dimethylsiloxane) (PDMS) stamp printing and high-precision pin printing, both of which are described in the following sections.

3.4.

Steady-State Characterization Setup

To achieve accurate temperature calibration of the sensors produced for this application, it was necessary to characterize their response over the relevant oxygen concentration range at a series of different gas temperatures. A gas delivery setup comprising mass flow controllers (MFCs, Celerity, Ireland) for oxygen and nitrogen (the carrier gas), and an air process heater (Radionics, Ireland) was employed to this end. The MFCs were controlled using custom-written LabVIEW software (National Instruments, United Kingdom), and the gas temperature was varied by applying a range of different voltages to the air process heater. For the purposes of a typical sensor calibration, the optical module was placed in a custom-made gas cell, which was connected to the gas delivery setup. The temperature within the cell was recorded using a thermistor (BetaTherm, Ireland), the data from which were transmitted along a separate input channel of the detection electronics.

A similar experimental arrangement was employed when characterizing the humidity sensitivity of the sensors, the primary difference being the inclusion of water-filled, gas washbottles (AGB Scientific, Ireland) in the gas delivery system. These served to humidify the sample gas flow. All sensor data (including temperature) were recorded using a custom-written LabVIEW program.

3.5.

Breath Simulation Apparatus

The ability to characterize sensor response across a broad range of breath rates in a reproducible manner was an integral aspect of the work reported here. A commercially available metabolic calibration unit (VacuMed, Ventura, California) was employed to this end. This device is essentially a lung simulator, which mimics oxygen consumption through the dilution of an inspired volume of ambient air with a mixture of carbon dioxide in nitrogen (referred to as the calibration gas). The flow of calibration gas is controlled using a needle valve and monitored using a digital mass flow meter. Breathing is simulated by a motorized piston, which draws ambient air plus the calibration gas into a drum, and then expels the resultant mixture. The breath rate is user adjustable and can be varied from approximately 8 breaths per minute to 60 breaths per minute.

For this work, the sensor module was attached to the outlet of the lung simulator and its response to changes in oxygen concentration was recorded over a range of breath rates using the LabVIEW software mentioned in the previous section.

4.1.

Steady-State Calibration of Stamp-Printed Sensors

The steady-state responses of stamp-printed sensors were recorded using the experimental setup described in Sec. 3.4. Sensors were exposed to oxygen concentrations ranging from 15 to 23%, which defines the range of interest for this application, and the response of a typical sample for four different gas temperatures is shown in Fig. 2 . The performance of the sensor module can be characterized by a large signal-to-noise ratio $(\mathrm{SNR}>1100)$ and a high degree of response stability and repeatability. As the temperature increases, a reduction in sensitivity is apparent, along with a downward shift in the response baseline. These and similar data were subsequently used to yield the 3-D calibration surface shown in Fig. 3 (generated using Table Curve 3D, Systat Software Incorporated, San Jose, California), which is described by the third-order polynomial equation $(R^2=0.9995)$

Fig. 2

Steady-state response of stamp-printed oxygen sensor for a range of temperatures.

Fig. 3

Calibration surface for a typical stamp-printed sensor.

4.2.

Effect of Humidity on Sensor Performance

Having demonstrated the ability of the sensor to respond effectively over the ${\mathrm{O}}_2$ concentration range of interest, it was necessary to investigate the degree to which its performance was affected by changes in ambient humidity. The results of these experiments would have a significant influence in determining the suitability of this device for breath monitoring applications.

A typical sensor was exposed to both dry and humid mixtures of ${\mathrm{O}}_2$ in ${\mathrm{N}}_2$ , as described in Sec. 3.4. Figure 4 illustrates the calibration curves of the sensor when exposed to both dry (◼) and humid $(●)$ gas flows. Here, the term “dry” refers to gas having a RH value of less than 5%, while “humid” implies a RH value of greater than 95%. It is clear that, for these two extremes, the sensor calibration curves differ minimally, demonstrating the high degree of RH independence displayed by the doped host matrix for ${\mathrm{O}}_2$ concentrations ranging from 0 to 100%. This is an extremely significant result in the context of developing a breath gas analyzer, as it facilitates the development of a system that does not require the integration of gas drying or sample heating strategies into its design to eliminate the effects of humid breath.

Fig. 4

Calibration curves of a typical PTEOS-derived sensor for both dry (◼) and humid $(●)$ gas flows. Inset shows step-by-step response of the sensor to varying ${\mathrm{O}}_2$ concentrations.

These preliminary experiments demonstrated the obvious potential of this sol-gel-based sensor for the detection of oxygen in breath. The following stage in this work was the analysis of the sensor performance on a breath-by-breath basis.

4.3.

Breath-by-Breath Performance

The sensor module was attached to the output of the lung simulator, and its performance was characterized at a breathing rate of ten breaths per minutes for a 2% change in oxygen concentration. The sensor output was then calibrated using the third-order polynomial described in Sec. 4.1. The resultant, temperature-calibrated response is shown in Fig. 5 . Here, the ability of the sensor to respond to changes in ${\mathrm{O}}_2$ concentrations on a breath-by-breath basis is clearly demonstrated, but the true strength of the device lies in its ability to record the in-breath ${\mathrm{O}}_2$ profile, a characteristic that is lacking in commercial analyzers. The sensor displays a resolution of approximately 0.03% ${\mathrm{O}}_2$ and clearly detects the expected 2% change in ${\mathrm{O}}_2$ concentration.

Fig. 5

Temperature-calibrated response of stamp-printed sensor to simulated breath.

However, closer examination of the data indicates an apparent limitation of the sensor response time, as evidenced by the shape of the response curve for the higher oxygen concentration. This corresponds to the inspiratory cycle of the simulator, where the sensor spot is being exposed to ambient air. The absence of a stable baseline, or plateau in the sensor output, is indicative of incomplete diffusion of oxygen into the sol-gel sensor matrix before the subsequent expiratory cycle of the simulator. This effect is even more apparent for the expiratory cycle of sensor response, i.e., the cycle corresponding to the diffusion of ${\mathrm{O}}_2$ molecules out of the sensor matrix. This is consistent with the response and recovery characteristics of such sol-gel films, and would appear to suggest that the sensor response time is inadequate at this breathing rate. Given that the response shown was recorded at the relatively low breathing rate of 10 breaths per minute, it was decided to examine the sensor performance in terms of the output signal change at higher breathing rates to more fully understand the extent of this limitation.

4.3.1.

Investigation of effective measurement range

Stamp-printed sensor

Figures 6a and 6b display the response of a stamp-printed sensor to a 2% change in oxygen concentration at breathing rates of 10 and 30 breaths per minute, respectively. For the purposes of clarity, it is important to note that the sensors were not temperature calibrated for the purposes of these experiments, as this was not necessary to examine their response range. The data are, therefore, displayed in terms of the phase angle produced by the sensor in response to oxygen. This has the effect of inverting the waveform when compared to that displayed in Fig. 6, with lower phase angles corresponding to the inspiratory phase of the simulator (i.e., the reduced lifetime/higher quenching regime).

Fig. 6

Response of stamp-printed sensor at breath rates of (a) 10 breaths and (b) 30 breaths per minute.

There is clearly a significant difference between the data recorded for the two breathing rates, with the sensor registering a reduction in overall signal change of $0.09\mathrm{deg}$ when exposed to the higher breathing rate. This has obvious implications for effective sensor performance, as such a result demonstrates an apparent loss in sensitivity at higher breathing rates. This limitation was thought to be due to the thickness of the sensor spot, a parameter that is intrinsically linked to its response time. When such stamp-printed spots were examined by optical profilometry (Wyko optical profilometer, Veeco Limited, United Kingdom), they were found to be 35 to $45\mathrm{\mu }\mathrm{m}$ in thickness, which is a direct result of the PDMS stamp-printing technique and the relatively large volume of material that is deposited. Ideally, sensor spots should be less than $1\mathrm{\mu }\mathrm{m}$ thick to achieve the subsecond response times that are characteristic of such sol-gel-based, oxygen sensors.⁵ Clearly, the technique of PDMS stamp printing was not compatible with this criterion, and high-resolution pin printing was therefore examined as an alternative. It was expected that this technique would not only yield thinner sensor spots due to the reduced volume of deposited material, but that its automated nature would lead to a significant improvement in the reproducibility of sensor spot deposition.

Pin-printed sensor

Oxygen-sensitive spots with a diameter of approximately $100\mathrm{\mu }\mathrm{m}$ were deposited onto the sensor chips by pin printing, as described in Sec. 3.3.2. Each sample that was investigated consisted of a single sensor spot per frustrated cone. The response of a typical pin-printed chip to a 2% change in ${\mathrm{O}}_2$ concentration is shown in Figs. 7a and 7b for breathing rates of 10 and 30 breaths per minute, respectively. There is a clear improvement in response compared with that of a stamp-printed sample. This is evident in the shape of the waveform, which plateaus during the inspiratory cycle, this being indicative of complete diffusion of oxygen into the sensor film before the expiratory cycle of the simulator. The improvement in response is further reinforced on examination of the apparent sensitivity of the pin-printed sensor at higher breathing rates. The signal change at 30 breaths per minute has reduced by only $0.03\mathrm{deg}$ , i.e., one third of the reduction recorded for the stamp-printed sensor. On examination by optical profilomentry, pin-printed sensor spots were found to be 3 to $4\mathrm{\mu }\mathrm{m}$ thick, which, as expected, yielded a much improved performance.

Fig. 7

Response of pin-printed sensor at breath rates of (a) 10 breaths and (b) 30 breaths per minute.

Figure 8 illustrates more clearly the relative performances of the stamp- and pin-printed sensors for breathing rates ranging from 10 to 50 breaths per minute. The stamp-printed sensor displays a 40% apparent loss in sensitivity, compared with a loss of just 8% for the pin-printed sensor across the entire range.

Fig. 8

Comparison of breath rate dependence for stamp-printed (◼) and pin-printed $(●)$ sensors.

In addition, the pin-printed sensor achieves a resolution of 0.03% ${\mathrm{O}}_2$ for a 2% change at 10 breaths per minute, demonstrating its ability to not only outperform the stamp-printed sensor in terms of its effective measurement range, but to match its sensitivity at low breathing rates. Such a resolution compares favorably with the 0.05% (vol) quoted for the Oxycon Mobile system supplied by Viasys Healthcare Incorporated (Conshohocken, Pennsylvania), which is the most portable, commercial metabolic gas analyzer that is available at present.¹³ This performance, combined with the additional ability to record the in-breath oxygen profile, emphasizes the potential of the sensor reported here.