Abstract:
Automated severity assessment of disease in computed tomography (CT) images plays an essential role for regional assessment of disease that is in great need of liver-related computer assisted diagnosis. An effective dataset is a cornerstone of the deep learning-based method. However, the popular liver-related datasets only focus on specific tasks and fail to guide advanced liver diagnosis tasks. To bridge the gap, we construct the first publicly comprehensive liver dataset, called ComLiver, which consists of multiple liver-related tasks and manually marks the elaborate labels for each task (500 cases). Note that the images in the ComLiver dataset are all thin thickness data that have significant clinical implications. In addition, liver and liver lesions segmentations have attracted substantial interest and achieved approving progress, other liver-related tasks of the liver are still under-studied due to various challenges (eg low contrast, vascular complexity, etc.), despite its significance in assisting preoperative planning. To better exploit the advanced tasks in liver therapy, we introduce vessel instance segmentation, couinaud segmentation tasks based on previous tasks. Finally, we perform a thorough evaluation of the state-of-the-art methods of each task on the proposed ComLiver dataset and obtain a number of interesting findings. Results show its challenging nature, unique attributes and present definite prospects for novel, adaptive, and generalized liver-related segmentation methods. We hope this dataset could advance research towards liver-related computer assisted diagnosis.
