Preprocessing and Semantic Segmentation of Medical Images of Lung Diseases using AI Techniques

Abstract

Automated lung region segmentation in computed tomography (CT) images is vital for diagnosing and treating lung diseases. This research leverages artificial intelligence (AI) techniques to enhance the preprocessing and semantic segmentation of lung disease images. Two approaches are proposed: one integrating YOLOv8 and U-Net, and the other combining YOLOv8 with Attention U-Net for improved segmentation precision. These approaches efficiently identify and segment lung regions from CT images, enhancing accuracy and robustness. Using a dataset of 140 CT exams, the approaches were trained and evaluated using metrics such as Dice coefficient, IoU, precision, recall, and computational cost. The first approach showed significant improvements, with a training Dice Coefficient reaching 0.9900 and a validation Dice Coefficient of 0.9788. The second approach also demonstrated robust performance, with validation precision reaching about 0.996 and recall about 0.987. Additionally, a web application called Lung Vision, which facilitates medical image management, is introduced. The results underscore the high accuracy and efficiency of the proposed approaches, offering innovative solutions for lung segmentation and practical applications in clinical settings, thereby advancing patient care through advanced AI technologies.