Lightweight network improves medical image segmentation accuracy and speed
GAD-MambaUNet: Direction-Group Mamba with Gradient-Adaptive DINOv3 Distillation for Lightweight Medical Image Segmentation
Computer Vision and Pattern Recognition
Summary
Medical image segmentation helps computers find important parts in scans like MRIs quickly and accurately. The authors created a small but smart network called GAD-MambaUNet that looks at scan directions and groups of image features together to understand images better. They also used a powerful pre-trained model to teach the network during training without making it bigger. This approach balances being fast and accurate, making it useful for medical imaging tasks where computing resources are limited.
What this means in practice
- •For medical imaging engineers: Deploy compact segmentation models that efficiently analyze medical scans with improved accuracy on resource-limited hardware.
- •For mobile health software developers: Integrate lightweight and accurate segmentation tools into mobile health apps for real-time medical image analysis.
Authors
Fang Wang, Huitao Li, Wenhan Chao, Zheng Zhuo, Xinxin Yang
Abstract
In this paper, we proposed GAD-MambaUNet, a lightweight medical image segmentation network that combines efficient local modeling, direction--group state-space interaction, and training-time foundation-model supervision. To improve contextual modeling in compact segmentation networks, we introduced Direction-Group Graph Selective Scan (DG-GSS), which treated scan-direction and channel-group responses as graph nodes and enabled structured information exchange before multi-directional fusion. We further incorporated DINOv3-GAD supervision, where a frozen DINOv3 teacher provided semantic guidance during training, and Gradient-Adaptive Distillation dynamically regulated the distillation strength. GAD-MambaUNet achieves a favorable accuracy--efficiency balance compared with representative lightweight and general segmentation methods. Ablation studies further verify the effectiveness of DG-GSS and training-time DINOv3-GAD supervision. In future work, we will explore more flexible teacher--student alignment strategies and extend the proposed framework to more diverse medical segmentation scenarios, such as multi-class and multi-modal segmentation tasks.