Pneumonia detection with deep learning approaches from numerical data and chest X-ray images

Pneumonia is a lung infection with high morbidity and mortality in all age groups worldwide. Effective detection of the disease in the early stages is crucial to reduce pneumonia-related deaths and increase cure rates. Techniques such as physical examination, laboratory tests, clinical symptoms and various medical imaging methods are used to identify pneumonia. Because other lung diseases present similar symptoms to pneumonia and imaging findings are non-specific, the diagnosis of pneumonia is time-consuming and error-prone. The adoption of deep learning-based methods to assist experts in the early diagnosis of diseases is achieving effective results as an accurate and efficient diagnostic method in medical applications, including pneumonia diagnosis. The aim of this thesis is to diagnose pneumonia from numerical data and chest X-ray (CXR) images using deep learning methods. Machine learning and deep learning algorithms were applied to a numerical medical dataset of 2000 individuals (1000 pneumonia, 1000 healthy) with demographic characteristics, symptoms and laboratory test results. CXR images of the same individuals were also classified using deep learning-based methods. The classification results of the numerical data and CXR images were compared. The data of the study were collected by retrospectively reviewing the files of patients admitted to the Chest Diseases and Tuberculosis clinic, intensive care unit and chest outpatient clinic of Dicle University Faculty of Medicine. One of the aims of this study is to develop and improve the multidimensional diagnosis of pneumonia with a deep learning-based computer-aided automated detection system that will help experts where there is a lack of expensive equipment and highly trained clinicians. When the results obtained in this thesis are compared with other studies, it is found that the proposed methods contribute to the existing literature and perform well.