MsM-DPM: Multiscale Mamba Diffusion Probabilistic Model for Medical Image Segmentation.

Area of Science:

• Medical Image Analysis
• Deep Learning
• Computer Vision

Background:

• Traditional diffusion probabilistic models (DPMs) struggle with irregular medical image structures and lesion-background similarity.
• Accurate medical image segmentation is crucial for diagnosis and treatment planning.

Purpose of the Study:

• To propose an innovative multiscale Mamba DPM (MsM-DPM) architecture for enhanced medical image segmentation.
• To improve robustness to shape and scale variations in lesions.
• To better capture lesion boundary details and semantic differences.

Main Methods:

• Developed MsM-DPM integrating a multiscale attention fusion module (MSAFM) and a multiscale denoising UNet (Ms-DU).
• Incorporated a multilayer axial feature module (MLAFM) for adaptive global context aggregation.
• Utilized a multilevel global context (MLGC) module for skip connection reconstruction and a feature fusion module (FFM) for boundary detail enhancement.

Main Results:

• MsM-DPM demonstrated superior performance across six diverse datasets (LUNA16, ATM22, COVID-19, Self-collected, Pancreas, BT-MSD).
• The architecture effectively encodes semantic differences, improving feature representation for lesions.
• Outperformed existing medical image segmentation methods in experimental evaluations.

Conclusions:

• The proposed MsM-DPM offers a significant advancement in medical image segmentation accuracy and robustness.
• The novel architecture effectively addresses limitations of traditional DPMs in complex medical imaging scenarios.
• MsM-DPM shows promise for clinical applications requiring precise lesion segmentation.