State-of-the-Art Deep Learning Method and Its Explainability for Computerized Tomography Image Segmentation

Medical image segmentation is an important task for computer-aided diagnosis. It provides information on the target structure for physicians to perform accurate diagnosis and treatment. Recent advances in deep learning enable the possibility of fully automatic image segmentation. In this chapter, we first review the technical aspects of the deep learning method, including data preprocessing, model development, and implementation and training strategy. Two clinical applications are used to demonstrate the potential of the deep learning method: (1) aorta and coronary artery segmentation on cardiac computerized tomography coronary angiography (CTCA) and (2) airway segmentation on a high-resolution computerized tomography (HRCT) image. Then current performance evaluation methods are discussed. Furthermore, the explainability of deep learning methods is essential for humans to understand the principles and mechanisms of CT image segmentation. Therefore, the techniques and tools in the area of explainable AI (XAI) are introduced and highlighted. Finally, limitations and future opportunities are addressed.