Medical Image Segmentation based on Deep Active Contour and Mean Curvature Loss Function

To address the issue of lack of geometric prior information in existing pixel-level training, researchers proposed a medical image segmentation method that combines deep active contours with average curvature loss functions. This method integrates the Chan-Vese model into the loss function and introduces average curvature as a geometric natural constraint, using convolution kernels for approximate calculations to save computational resources. The research team verified the performance of this method on liver and spleen datasets, and the results showed that it reached new advanced levels on multiple segmentation datasets.