RWKVFUSE: Parallel Encoders Network for Medical Image Segmentation

Medical image segmentation is essential for accurate lesion identification and disease diagnosis. Conventional CNN-based methods are limited by their restricted receptive fields, whereas Transformer models effectively capture global dependencies but incur quadratic computational costs. Recently, sequence modeling approaches with linear attention mechanisms, such as the RWKV, have demonstrated the ability to model long-range dependencies with linear complexity. To address this, we propose a dual-encoder architecture, RWKVFuse, which integrates VRWKV and CNNs to extract fine-grained spatial features and long-range dependencies. Extensive experiments across four imaging modalities show that the proposed RWKVFuse model outperforms state-of-the-art methods in metrics such as DSC and HD95. Our code is available at: https://github.com/alqurri/rwkv