Breaking the Disconnection Dilemma in Class-Imbalanced Cerebrovascular Segmentation

Cerebrovascular segmentation plays a vital role in medical image processing, as it aids in diagnosing and treating cere-brovascular diseases. However, current methods often fail to meet the practical accuracy standards required in clinical settings. This paper presents a novel cerebrovascular segmentation method based on 3D UNet. First, we construct a connectivity matrix to obtain cerebrovascular connectivity information, which is then combined with the centreline distance weighted loss to effectively reduce errors caused by disconnection and class imbalance. To validate the effectiveness of this method, we conducted a series of experiments on a public dataset, and the results demonstrated significant performance improvements across multiple metrics in the cerebrovascular segmentation task.