LightVM-SparseUNet: a lightweight medical image segmentation framework via Vision Mamba and sparse attention

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Traditional Convolutional Neural Networks (CNNs) and Transformer architectures face challenges with parameter redundancy and computational complexity.
• These limitations hinder the deployment of medical image segmentation models on resource-constrained edge devices.

Purpose of the Study:

• To propose LightVM-SparseUNet, an ultra-lightweight medical image segmentation framework.
• To address the deployment challenges of medical imaging AI on edge devices.

Main Methods:

• Developed a Multi-path Visual Mamba (MVM) module for efficient feature extraction with linear complexity.
• Integrated a Sparse-Sampling Self-Attention (SSSA) mechanism into U-shaped skip connections to capture long-range spatial dependencies efficiently.

Main Results:

• LightVM-SparseUNet achieves segmentation performance competitive with state-of-the-art large-scale models.
• The model demonstrates extreme lightweight design with only 0.08 million parameters and 0.16 GFLOPs computational overhead.
• Experimental validation on two public datasets confirms the model's efficacy.

Conclusions:

• LightVM-SparseUNet offers a highly practical and efficient solution for medical image segmentation on edge devices.
• The proposed framework significantly reduces computational complexity and parameter count while maintaining high performance.