WCA-net: A Medical Image Segmentation Network via Wavelet Learning with Coordinated Calibration Attention
Zenghui Li, Changming Song, Dongxu Cheng · 2024
Automatic segmentation is essential for diagnosis and cure-planning in the medical image field. Nowadays, many deep-learning methods have been proposed to segment anatomy regions in medical image segmentation and have shown excellent performance. There still exist some issues. We proposed a novel medical image segmentation network via wavelet learning with coordinate calibration attention. Firstly, the DRF (Dual Receptive Field Branch) combines convolution block and wavelet learning to capture more semantic information. Secondly, the CCA (Coordinate Calibration Attention) module addresses the feature misalignment problem that arises when fusing contextual information at different scales. MDSF (Multi-scale Deep Supervision Fusion) obtains the output by integrating the multi-scale features and applies supervision during network training to achieve improved performance. The results obtained from the ISIC2016 and 3DIRCADb datasets demonstrate the competitive performance of our proposed method.