Pulmonary tuberculosis is an infectious disease caused by Mycobacterium tuberculosis complex and transmitted mainly by airborne droplets, the development of which in infected people is associated with the presence of unfavorable risk factors that increase during periods of military action, therefore the development of rapid diagnostic methods for tuberculosis that are available for use in field conditions is relevant.

OBJECTIVE

To evaluate the possibility of differential express diagnostics of pulmonary tuberculosis based on the composition of exhaled air using a portable device of the "electronic nose" class of adults and children, including those with an aggravated anamnesis.

MATERIAL AND METHODS

Samples of exhaled air were collected from 98 people aged 11—92 years, including 55 patients with tuberculosis and 43 healthy people, including 53 smokers and 37 people with concomitant chronic diseases, using a portable device "electronic nose" developed by us based on an array of 8 metal-oxide sensors supplemented with temperature and humidity sensors. The array of response data collected according to the developed algorithm was divided into two non-overlapping samples: training (68 people, including 37 patients with tuberculosis) and test (30 people, including 18 patients with tuberculosis). The

training set was used for machine learning by the linear regression method, the training efficiency was assessed in a blind test using the test set.

RESULTS

The sensitivity and specificity of the developed approach in cross-validation of the training set were 84% (6 false negative (FN) results, (CI 95%: 75%—93%)) and 71% (9 false positive (FP) results, (CI 95%: 59%—83%)), respectively. At the same time, for healthy people, the specificity was 87%, and for study participants with comorbidities — 56%. The sensitivity and specificity of the

developed approach, achieved in a blind test, were 83% (3 FN results, (CI 95%: 73%—93%)) and 92% (1 FP result, (CI 95%: 85%—99%)). Conclusions: The fundamental possibility of non-invasive express diagnostics of tuberculosis, both with and without bacilli excretion, based on the composition of exhaled air in a time not exceeding 5 minutes has been demonstrated. The use of a portable device "electronic nose" based on an array of 8 metal oxide sensors supplemented with temperature and humidity sensors, using the developed algorithm for collecting exhalation and subsequent automated data analysis, made it possible to achieve

sensitivity and specificity of 83 and 92%, respectively, in a blind test. The disadvantage of the developed approach remains poor differential diagnostics of tuberculosis with other lung diseases (non-tuberculous mycobacteriosis, lung cancer, asthma), which requires further research.