The paper introduces a new shape-aware loss function called FourierLoss, used in encoder-decoder networks for various medical image segmentation tasks. This function quantifies the shape dissimilarity between the ground truth and the predicted segmentation maps, penalizing this dissimilarity in network training. FourierLoss offers an adaptive loss function with trainable hyperparameters that control the importance of the shape details the network is enforced to learn. The paper presents experiments on 2879 computed tomography images of 93 subjects, suggesting that FourierLoss significantly improves results for liver segmentation.
Publication date: 22 Sep 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2309.12106
