Medical Image Segmentation via Sparse Coding Decoder

Transformers have achieved significant success in medical image segmentation, owing to their capability to capture long-range dependencies. Previous studies have employed either pure Transformer or hybrid CNN-Transformer architectures in the encoder module to enhance their ability to extract more complex features. However, these models still exhibit limitations in fine-grained local feature extraction and effectively suppressing irrelevant information. To address this issue, a convolution sparse vector coding-based decoder is proposed, namely the CAScaded multi-layer Convolutional Sparse vector Coding DEcoder (CASCSCDE), which suppresses noise by refining the feature representation to be more sparse and accurate through sparse coding and localized convolution, effectively minimizing less important, noisy components. To demonstrate the effectiveness and versatility of our CASCSCDE, we incorporate our decoder into both pure Transformer and hybrid CNN-Transformer models, such as SwinUNet and TransUNet. Our experiments demonstrate that integrating CASCSCDE into the models significantly enhances segmentation performance. The CASCSCDE opens new ways for constructing decoders based on convolutional sparse vector coding.