LTMSegnet: Lightweight multi-scale medical image segmentation combining Transformer and MLP

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Medical image segmentation is crucial for research and diagnosis.
• Current neural network methods offer high accuracy but are computationally intensive for mobile devices.
• Lightweight models are needed for efficient deployment on low-resource hardware.

Purpose of the Study:

• To propose a lightweight medical image segmentation method.
• To achieve accurate segmentation with reduced computational cost for mobile applications.
• To integrate Transformer and Multi-Layer Perceptron (MLP) architectures.

Main Methods:

• Developed a U-shaped network integrating three novel modules: Multi-scale Branches Aggregate (MBA), Lightweight Shift MLP (LSM), and Feature Information Share (FIS).
• MBA module aggregates global spatial and local details for accurate feature learning.
• LSM module uses shift operations for inter-pixel associations, and FIS module refines feature fusion via skip connections.

Main Results:

• The proposed method was validated on ISIC 2018 and 2018 DSB datasets.
• Demonstrated superior performance compared to existing state-of-the-art lightweight segmentation methods.
• Achieved a significant balance between segmentation accuracy and computational efficiency.

Conclusions:

• The novel lightweight segmentation method effectively addresses computational challenges for mobile deployment.
• The integration of Transformer and MLP architectures offers a promising direction for efficient medical image analysis.
• The method provides a practical solution for accurate medical image segmentation on resource-constrained devices.