FastSCVM: A lightweight hybrid Mamba–convolution network for efficient glaucoma image segmentation

Glaucoma is a chronic eye disease that slowly damages the optic nerve and leads to vision loss. So, early screening and accurate segmentation of the affected areas are very important for timely treatment. Modern segmentation methods usually use CNNs or Transformer-based models. But, CNNs are limited by their local view, and Transformers need a lot of computing power. To address these challenges, we propose FastSCVM, a fast and lightweight segmentation framework that combines multi-scale convolutional operations with Vision Mamba modules. This combination enables the model to capture long-range dependencies while preserving lower computational overhead. Also, we introduce a Boundary-aware Mamba Fusion module (BMF) to enhance edge detection. Experiments on four publicly available glaucoma datasets—Drishti-GS, REFUGE, RIM-ONE-r3, and RIM-ONE_DL—demonstrate that the proposed method achieves strong segmentation performance, with Dice coefficients of 0.9141, 0.8857, 0.8623, and 0.8847 for the optic cup, respectively. Notably, the model requires only 0.6 million parameters and approximately 1.055 GFLOPs. The code is available at https://github.com/Xtaotianxiawudi/FastSCVM .