Disease detection and classification from lung images using deep learning method

Lung diseases are among the most important health issues that need to be addressed in the healthcare field. Early detection of the disease and a rapid response to the correct treatment are of vital importance. Doctors' ability to identify the disease at an early stage and the early initiation of treatment in affected individuals have significantly influenced the course of the disease. The detection and classification of lung diseases is a difficult process requiring expertise, and the timely implementation of the necessary intervention determines the success of the treatment. Lung diseases are diversifying every day and have become difficult to analyze. Incorrect disease detection and failure to analyze the disease correctly have greatly affected human health. In today's studies, the use of artificial intelligence methods is at the forefront in detecting disease. Analyzing images and classifying them correctly alleviates the condition of the disease and reduces the risk of death. In this thesis, a dataset consisting of bacterial, viral, and normal disease images from lung images was used with YOLO models, and different versions of the YOLO algorithm (YOLOv8, YOLOv9, YOLOv10, YOLOv11, YOLOv12) were examined comparatively with the aim of detecting disease. In the experiments, YOLOv8 achieved an average precision (mAP) value of 87.9%, YOLOv9 achieved 85%, YOLOv10 achieved 86.8%, YOLOv11 achieved 87.8%, and YOLOv12 achieved 86.1%. In the study, YOLOv8 showed the best performance compared to other models due to its ability to generalize object detection thanks to its model architecture and feature extraction. To improve the disease detection capability of the study, the SE and attention modules were integrated into the YOLOv8 model, which performed the best. The developed YOLOv8 model achieved the best performance with 88.7% mAP, 88% sensitivity, 78.6% precision, and 83% F1 score. The SE and attention module integrated into the YOLOv8 model improved the study's result by 1.01%.