Development of deep learning model for detection of lung diseases from tomography images

In recent years, Health Information Systems have shown significant developments in our country. These indicators have gained momentum in medical informatics with Hospital Information Management Systems (HIS) and Picture Archiving and Communication System (PACS). From a patient's admission to the hospital until their discharge, all processes are digitally tracked, emphasizing the importance of data in artificial intelligence studies. Radiological deductions made through medical imaging methods play an essential role in diagnosis. The Picture Archiving and Communication System (PACS) is used nationally and internationally for transmitting and archiving medical images. In this study conducted at a hospital in Turkey, lung tomography image data from the Picture Archiving and Communication System (PACS) were used. Early diagnosis of multiple diseases observed on lungs has become critical during the COVID-19 pandemic. In this study, segmentation and classification work using deep learning methods were carried out for diagnosing pneumonia among lung diseases. In the study, honeycomb cells seen on diseased lungs' surfaces and ground-glass areas formed the basis for diagnosing pneumonia. In this study, an original deep learning model was developed using Convolutional Neural Networks for image classification tasks. The deep learning model created was compared with models built using architectures of convolutional neural networks on the same dataset. A review of existing literature on studies conducted with lung x-ray images and lung tomography images was conducted. Pneumonia and healthy lung tomography images were divided into three classes: training, testing, and validation. The trained model achieved an accuracy of 98.16% on the training data, 97.57% on the validation data, and 95.19% on previously unseen test data.