1.

Introduction

Noninvasive diagnostics are one of the most important directions for the development of modern medicine. Recently, the interest has been focused on patients’ exhaled air research as a noninvasive diagnostic method for bronchopulmonary, cardiovascular, gastrointestinal, and other diseases.^1,2 The basis of similar methods is related to the variation of concentrations of volatile organic compounds (VOCs) in exhaled air according to specific diseases.³ For example, it is ascertained that bronchial asthma exacerbation is characterized by ammonia (${\mathrm{NH}}_3$) concentration increasing in exhaled air by 250 to 300 times.^3,4 A high level of propane (${\mathrm{C}}_3{\mathrm{H}}_8$) in the exhaled air was identified for different clinical forms of pulmonary tuberculosis.^5,6 The issue is significant because this approach, due to the sparse sampling, precludes pain and physical and emotional discomfort of the patient, the possibility of transmission of the blood-borne infections, and provides safety for the diagnostic studies. On one hand, noninvasive diagnostic methods can be used on an outpatient basis—this provides their widespread application; on the other hand, for patients at resuscitation departments, as the severity of the patient’s state is not a contraindication for their application.

In this paper, we discuss the abilities of the methods of infrared (IR) laser spectroscopy and the principal component analysis (PCA) technique for a totally noninvasive express diagnostic of chronic obstructive pulmonary disease (COPD) on the basis of absorption spectra analysis of the patient’s exhaled air.

2.

Techniques and Methods

The method of laser photoacoustic spectroscopy (LPAS) is convenient for measuring the concentration of VOCs in the exhaled air because of the simplicity of its practical implementation, safety, cost-effectiveness, and extremely high sensitivity (minimal measured concentration for some chemicals at atmospheric conditions is about 1 ppb). LPAS has advantages in the detection of the gases which have overlapping of the absorption lines with the lasing lines (e.g., CO or ${\mathrm{CO}}_2$-laser). LPAS is based on the generation of acoustic waves in a gas excited by the modulated laser beam at the wavelength corresponding to the absorption line of the VOCs in gaseous samples and on the measurement of the parameters of these acoustic waves using sensitive microphones.

3.

Results and Discussion

The experimental research was carried out according to the principles of Good Clinical Practices. The protocol of the research was approved by the Ethic Committee of the Siberian State Medical University (Tomsk, Russia), Ref. Number 2882 on 24 November 2011. All participants were informed about details of the research and signed “Informed agreement” for the actions carried out. The study involved 11 healthy nonsmoking volunteers (control group) and seven patients with COPD (target group). The COPD patients were males with verified diagnoses who passed treatment at the Pulmonological Division of the Regional State Autonomous Institution of Public Health “Municipal Clinical Hospital No. 3” (Tomsk, Russia). The average age of this group was 59.6 years. We did not included COPD patients with an unverified diagnosis, the presence of pneumonia, asthma, and other respiratory pathologies. The control group consisted of conventionally healthy nonsmoking male volunteers. Inclusion criteria were the absence of acute illness within two weeks prior to sample collection, without chronic pathologies of bronchopulmonary, cardiovascular, digestive, urinary and reproductive systems, and the absence of the factor “smoking” in anamnesis vitae. The average age in this group was 21.1 years.

The procedure of exhaled air sampling was as follows. All samples were taken before eating or 2 h thereafter. The air was collected in standard test tubes. Prior to sampling, participants rinsed their mouth with running water. The study does not imply special cleaning of the oral cavity. Then the participant did some calm breaths through a sterile plastic tube into the test tube, which was then sealed with a sterile cotton wad.

All exhaled air samples were analyzed using the LGA-2 and the LaserBreeze LPAS gas analyzers. Five scans of the absorption spectrum of each sample were recorded and averaged to reduce random errors.

Most informative subranges of the measured profiles of the absorption spectra were determined by PCA. The criterion was the best spatial separation of the target group from the control group in the space of the PC. The results below are focused on the most informative subranges.

The number of PC (in other words, the dimensions of the above-mentioned space) is usually chosen to describe at least 70% of the variation of initial data [it is the so-called explained variance (EV)].¹¹ Here, initial data involved in the absorption spectra of all exhaled air samples. The value of EV in the used spectral subranges that is dependent on the quantity of the PC is presented in Table 4. According to the data from Table 4, the 2-D space of PC is enough to analyze profiles of the absorption spectra of the exhaled air samples.

Table 4

The dependence of the explained variance (EV) on the quantity of the used principal components (PCs).

To select the most informative set of absorption coefficients, we apply the method that is similar to the well-known “method of broken sticks” to the loading matrix.¹⁴ The results are shown in Table 5. Here, the initial quantity is the quantity of absorption coefficients which were contained in the definite spectral subrange before PCA application.

Table 5

The quantity of informative absorption coefficients.

According to the PCA, every sample is represented by the point in the space of PC. We used the freeware “ViDaExpert”¹⁵ to estimate the spatial distribution of the exhaled air samples in the space of the PC.

The results of point estimates of the exhaled air absorption spectra profiles of COPD patients and healthy volunteers in the 9.2 to $9.8\mu \mathrm{m}$ subrange are shown in Fig. 1. The distance between the point estimates on the plane of the PC characterized the difference in the absorption spectra profiles of participants. This is caused by variations in metabolism and, hence, is the difference in the VOCs profile of the samples.

Fig. 1

Spatial distribution of the exhaled air samples from chronic obstructive pulmonary disease (COPD) patients (the diamond icons) and healthy volunteers (the triangle icons). The feature set includes absorption coefficients of the sample in the range of 9.2 to $9.8\mu \mathrm{m}$. The axes correspond to the first (PC1) and the second (PC2) principal components.

The similar results of point estimates of the measured spectra of the exhaled air of COPD patients and healthy volunteers from the control group using gas analyzer LaserBreeze in the most informative subranges from 2.59 to $4.18\mu \mathrm{m}$ are shown in Figs. 2Fig. 3–4.

Fig. 2

Spatial distribution of the exhaled air samples from the COPD patients (the diamond icons) and healthy volunteers (the triangle icons). The feature set includes absorption coefficients of the sample in the range of 2.59 to $2.817\mu \mathrm{m}$. The axes correspond to the first (PC1) and the second (PC2) principal components.

Fig. 3

Spatial distribution of the exhaled air samples from the COPD patients (the diamond icons) and healthy volunteers (the triangle icons). The feature set includes absorption coefficients of the sample in the range of 3.272 to $3.498\mu \mathrm{m}$. The axes correspond to the first (PC1) and the second (PC2) principal components.

Fig. 4

Spatial distribution of the exhaled air samples from the COPD patients (the diamond icons) and healthy volunteers (the triangle icons). The feature set includes absorption coefficients of the sample in the range of 3.499 to $3.725\mu \mathrm{m}$. The axes are corresponded to the first (PC1) and the second (PC2) principal components.

Figures 2–4 show that the methods of IR LPAS and PCA allow separating patients with COPD and healthy nonsmoking volunteers.

Taking into account that the VOCs absorption bands correspond to the chosen spectral ranges, we can assume that spatial separation of the target and control groups is probably caused by the difference of hydrocarbon content in the COPD patients’ breath and the healthy nonsmoking volunteers’ breath. This matches to the results obtained by means of other methods for the analysis of the exhaled air.²

For further analysis, the absorption spectrum profile of a breath sample is suggested to be recognized as a “fingerprint” of the medical state of a patient. In order to estimate the possibility of diagnostics based on such “fingerprints,” the algorithm of soft*independent modeling of class analogy (SIMCA) was applied. SIMCA classification includes two stages.

The training stage. Each class of objects from the training set is independently modeled using PCA. In this result, the initial data are presented in the cloud form in the space of PCs. The coordinate origin is placed at the center of gravity of the cloud. Each object can be represented as the sum of two vectors: one lying in the cloud (projection) and another perpendicular to the first (residues). The average value (range) and deviation of the lengths of these vectors are the indicators belonging to this class of objects.

The testing stage. The classification procedure is as follows. Each new object is projected onto the built space (cloud). The obtained range and deviation are compared with the critical levels specified in the training stage.

In Table 6, we presented the examples of SIMCA classification using the profiles of the absorption spectra of breath samples in the range of 2.59 to $2.817\mu \mathrm{m}$ for various sets of samples for the training and testing stages.

Table 6

The classification of the absorption spectra of exhaled air of patients with COPD and healthy volunteers by SIMCA in the range of 2.59 to 2.817  μm.

The results in Table 6 show that analysis of the profile of exhaled air absorption spectra in the IR region allows us to separate COPD patients from the control group with a high enough accuracy.

4.

Conclusion

We described two types of laser photoacoustic gas analyzers which were developed by Special Technologies Ltd. for medical applications. Laboratory research of the exhaled air of patients with COPD and healthy nonsmoking volunteers was carried out at Siberian State Medical University (Russia) and in the Tomsk State University (Russia). The PCA method was used to select the most informative ranges of the absorption spectra of patients’ exhaled air in terms of the separation of the studied groups. It is shown that analysis of the profile of the exhaled air absorption spectrum allows identifying COPD patients in comparison to the control group. The most informative ranges of the absorption spectra of the COPD patients’ exhaled air and healthy nonsmoking volunteers’ exhaled air are 9.2 to 9.8, 2.59 to 2.817, and 3.272 to $3.725\mu \mathrm{m}$. The presented results are the base for the future construction of the classification rules for the noninvasive express diagnostics methods. There are two ways for classification rules construction. The first is to use the profile of the absorption spectrum of a breath sample as a “fingerprint” of the patient’s medical state. Another one consists of two steps: first, to carry out component analysis of breath samples for various groups, then to define the profile of the set of informative VOCs as a “fingerprint” of the patient’s medical state.