Radiuk P. Information technology for early diagnosis of pneumonia by the individual selection of parameters of the classification model by medical images of the lungs

Radiuk Pavlo. Information technology for early diagnosis of pneumonia by the individual selection of parameters of the classification model by medical images of the lungs. – Manuscript copyright. The thesis is on obtaining a scientific degree of Doctor of Philosophy in the field of knowledge 12 Information Technologies by specialty 122 – Computer Science. – Khmelnytskyi National University, Khmelnytskyi, 2021. The presented thesis is devoted to solving the current scientific and applied problem of automating the process of diagnosing viral pneumonia by medical images of the lungs through the development of information technology for the early diagnosis of pneumonia by the individual selection of medical parameters. It was defined in the dissertation that applying information technologies was highly topical in digital diagnostics of lung diseases on medical images of a thorax. According to the analysis of modern approaches, methods, and information technologies for diagnosing lung disease in the early stages based on medical images of the chest, the need to create information technology for the early diagnosis of pneumonia is substantiated. The scientific novelty and theoretical significance of the dissertation is that: 1) for the first time, a neural network model of a medical image of the lungs with the pneumonic inflammation features was created to identify pneumonia in the early stages; 2) the method of selection of quasi-optimal hyperparameters of the neural network model for identification of pneumonia in the early stages by X-ray images of the lungs was improved; the improved method of selection of quasi-optimal hyperparameters of the neural network model according to the convolutional architecture is intended for localization and detection of the scattered signs of pneumonic inflammation corresponding to individual features of lungs of the person; 3) the method of visual presentation and explanation of medical diagnosis was improved, which, unlike known approaches, is based on the formation of activation maps of classes based on weighted average gradients; 4) new information technology for early diagnosis of pneumonia by the individual selection of parameters of the classification model by medical images of the lungs was developed. The results of experimental testing of the proposed information technology have proven its ability to solve problems. Thus, the developed information technology surpassed the analogs in terms of classification accuracy by 0.58% and 1.95% in the CheXpert and PadChest datasets, respectively; according to the hit rate of 0.02% according to PadChest; completeness index of 0.69% and 1.64% for CheXpert and PadChest, respectively; according to the indicator of the area under the ROC-curve by 1.08% according to PadChest; by a second-error error rate of 0.64% and 1.64% for CheXpert and PadChest, respectively. High values of statistical indicators indicate both the high accuracy of detection of pneumonia and significant accuracy of the model in the identification of lung disease. The practical significance of the obtained results lies in software development for the early diagnosis of pneumonia based on medical images of the human chest. Using the information technology for early diagnosis of pneumonia allows: 1) performing effective identification of pneumonic inflammation on X-ray images of the lungs on small computing devices; 2) extracting the individual characteristics of each person's lungs on the medical image; 3) using a clear and straightforward user interface to detect mild pneumonia in the early stages of the disease. Thus, the results of experimental testing using the developed software confirm the accuracy of the scientific provisions of the proposed information technology, as the introduction of information technology can increase the reliability of detection of pneumonia by medical images by 0.58%-1.95% and reduce the probability of error during identification of lung disease by 0.64%-1.64% compared to known analogs. The results of the research were used in the educational process of Khmelnytskyi National University at the Department of Computer Science and Information Technology in the teaching of disciplines "System modeling," "Operations research and fundamentals of decision theory," "Intelligent data analysis," and "Applied mathematical packages for analysis of research results." The provisions developed in the dissertation have found practical application in the work of the radiology department of KP "Khmelnytskyi City Hospital," LLC "Scientific and Technical Firm" Infoservice" and in the educational process of Khmelnytskyi National University.