Wavelet Transform-based Distribution Discrepancy Maximization for Medical Image Segmentation

Accurate segmentation of organs and tumors is crucial for clinical diagnosis. Deep learning methods have been widely applied to various medical image segmentation tasks. However, these methods often suffer from the foreground and background class imbalance challenge when dealing with small regions of interest. To address this limitation, we propose a Wavelet Transform-based Distribution Discrepancy Maximization (WT-DDM) framework for medical image segmentation. Specifically, we first introduce a Distribution Discrepancy Maximization (DDM) module that makes the model on the foreground region from abundant irrelevant background. Then, the Wavelet Transform-based Feature Enhancement (WTFE) module was applied to mine the texture details of the foreground object. Experiments on multiple popular medical image segmentation datasets demonstrate that our framework yields highly competitive segmentation results.