MSCD-VM-UNet: A Vision Mamba Combining Multi-Scale Global and Local Feature Extraction With Cross-Domain Feature...

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate medical image segmentation is crucial for diagnosis and treatment planning.
• State Space Models (SSMs) offer linear complexity and long-range dependency modeling.
• Existing Mamba architectures struggle with multi-scale feature integration and boundary detail due to direct skip connections.

Purpose of the Study:

• To address limitations in current medical image segmentation models.
• To propose a novel architecture, MSCD-VM-UNet, for enhanced segmentation accuracy.
• To improve the integration of multi-scale features and handling of boundary details.

Main Methods:

• Developed the MSCD-VM-UNet architecture incorporating three novel modules.
• Implemented the Spatial Group Multi-Scale Attention Module (SGMAM) for multi-scale feature extraction and noise suppression.
• Integrated the Cross-Domain Feature Fusion Module (CDFFM) for frequency and spatial domain feature alignment.
• Utilized the Attention-Based Feature Injection Module (ABFIM) for adaptive feature fusion and weighting.

Main Results:

• The proposed modules significantly enhance the accuracy of the MSCD-VM-UNet architecture.
• MSCD-VM-UNet demonstrates superior performance in medical image segmentation tasks.
• The novel modules effectively capture multi-scale information and improve boundary detail handling.

Conclusions:

• MSCD-VM-UNet sets a new benchmark for medical image segmentation accuracy.
• The integration of SGMAM, CDFFM, and ABFIM modules overcomes limitations of previous architectures.
• This approach offers a robust solution for complex medical image segmentation challenges.