В даній роботі побудована нейромережа для діагностування захворювання на ХОЗЛ шляхом класифікації стану дихальної системи. Для діагностування використана розширена система діагностичних показників, що крім результатів загальноклінічного аналізу, анкетного опитування за визначеними стандартами та біохімічних показників включає також біофізичні показники. Біофізичні
показники є відсотковими внесками у складі конденсату вологи видихнутого повітря частинок різного походження та розміру. Аналіз
характеристик побудованої моделі включав значення точності, повноти та F-міри, які виявились найкращими при 30000 епохах
навчання нейромережі. Спроба використати ту ж саму нейромережу для вивчення стану дихальної системи після завершення лікування пацієнтів хворих на ХОЗЛ підтвердила, що у пацієнтів, які знаходяться на цій стадії, стан бронхо-легеневої системи не встигає
повністю відновитися.
Introduction. Neural networks are used to study and diagnose pulmonary diseases based on different sets of input data. In this work, a
neural network was built for diagnosing COPD by classifying the state of the respiratory system.
Material and methods. An extended system of diagnostic indicators is used for diagnosis, which, in addition to the results of a general
clinical analysis, a questionnaire based on defined standards, and biochemical indicators, also includes biophysical indicators. Biophysical
indicators are percentage contributions in the moisture condensate composition of exhaled air particles of different origins and sizes. The group
of patients consisted of 186 people aged from 40 to 45 years, patients with COPD in the exacerbation stage during inpatient treatment, during
(on the 5th day) and after treatment (on the 11-14th day) at the hospital stage, according to the existing protocol. During the examination,
in addition to standard methods of examination, samples of moisture condensate of exhaled air were obtained from patients using a special
device. The comparison group consisted of 373 practically healthy people aged 30 to 37 years. The values of all diagnostic indicators are
checked for quality before their use, cases of data lag in time, desynchronization, appearance of non-normalized data, omissions, irrelevance,
appearance of raw data and their duplication are monitored. When low-quality data is detected, appropriate technical and organizational
practices are applied to them, which aim to correct the situation. The obtained qualitative input data are divided into 3 groups: for training,
validation and testing of the neural network in a ratio of 3:1:1.
Conclusion. The neural network, which was built to diagnose COPD, showed high quality of the obtained solutions. Performance analysis
of the built model included accuracy, completeness, and F-measure values, which were found to be the best at 30,000 epochs of neural network
training. An attempt to use the same neural network to study the state of the respiratory system after the completion of treatment of patients
with COPD confirmed that in patients who are at this stage, the state of the bronchopulmonary system does not have time to fully recover. This
happens due to the complication of restoration of the homeostatic mechanisms of the functioning of respiratory tract tissues. Therefore, the
bronchopulmonary system of patients immediately after treatment cannot be considered completely healthy.
