A Multiscale Bidirectional Spiral Scanning State Space Mamba Network for Retinal Vessel Segmentation

Retinal vessel segmentation plays a critical role in the early screening and diagnosis of chronic diseases such as diabetic retinopathy and hypertensive vascular anomalies. However, traditional methods face persistent challenges due to the anatomical complexity of retinal vasculature, low-contrast imaging conditions, and interference from pathological artifacts like microaneurysms and exudates. Mamba models, leveraging State Space Models for linear-complexity sequence modeling, offer a promising alternative; however, their sequential scanning strategies often neglect spatial continuity and underperform in capturing high-frequency details at vessel termini. To address these challenges, this study proposes a hybrid architecture integrating convolutional neural networks (CNNs) with Mamba, named BISMUNet, which synergize local feature extraction and global contextual modeling. First, the encoder incorporates Vision Transformer inspired patch embedding to adapt Mamba modules, while the decoder retains CNNs to enhance fine vessel extraction. Then we design a bidirectional spiral Mamba block within the encoder, utilizing dual-directional spiral scanning paths to ensure vascular topological continuity. While we introduce a multiscale feature-enhanced attention module to dynamically fuse hierarchical features. Extensive experiments have been conducted on three publicly available retinal datasets (DRIVE, CHASEDB1, and STARE). Experimental results show that the proposed BISMUNet could achieve better segmentation results compared to other state-of-the-art methods. This work not only advances retinal vessel segmentation performance but also provides a scalable hybrid architecture paradigm for medical image analysis.