Multi-disease Detection and Segmentation of Chest CT Images Based on Coarse-to-Fine Pipeline Models

Accurate detection and segmentation of lung diseases are crucial for early diagnosis and treatment. In this study, we propose a multi-disease detection and segmentation method that utilizes a coarse-to-fine pipeline model, specifically targeting pneumonia and pneumothorax. Our approach consists of a lightweight U-Net for overall feature extraction and disease classification, followed by fine-tuning with the Segmentation Anything Module (SAM) for precise pixel-level segmentation in regions of interest. Attention mechanisms and skip connections are incorporated to enhance feature representation and contextual information propagation. Through extensive experimental evaluations on public multi-center datasets and private datasets, our method demonstrates superior performance in the detection and segmentation tasks of pneumonia and pneumothorax, outperforming other classic methods. These results highlight the potential of our approach as a rapid and accurate assistant tool for physicians, facilitating early diagnosis and personalized treatment of lung diseases.