CFG-MambaNet: Contextual and Frequency-Guided Mamba Network for medical image segmentation.

Area of Science:

• Medical Image Analysis
• Deep Learning
• Computer Vision

Background:

• Accurate medical image segmentation is crucial but challenging due to difficulties in global context modeling, boundary precision, and generalization.
• Existing methods, particularly Transformer-based models, face inefficiencies with high-resolution medical images and struggle with lesions exhibiting blurred contours or weak textures.

Purpose of the Study:

• To introduce a novel framework, CFG-MambaNet, designed to overcome the limitations of current medical image segmentation techniques.
• To enhance the efficiency, accuracy, and robustness of medical image segmentation, especially for challenging cases.

Main Methods:

• Employed a variable-scale state space block based on Mamba for efficient long-range dependency capture with linear complexity.
• Incorporated a frequency-guided representation module to distinguish low-frequency structures from high-frequency boundary details.
• Introduced an adaptive context aggregation mechanism for integrating semantic cues and highlighting critical regions, alongside a composite loss with deep supervision for stable training and improved boundary adherence.

Main Results:

• CFG-MambaNet demonstrated significant improvements in medical image segmentation across diverse datasets (ACDC, Kvasir-SEG, ISIC, SEED).
• The framework effectively addressed challenges related to long-range dependencies, boundary delineation, and segmentation of lesions with indistinct features.
• Achieved robust performance across various medical imaging modalities including cardiac MRI, endoscopy, dermoscopy, and pathology.

Conclusions:

• CFG-MambaNet offers an efficient and robust solution for medical image segmentation, outperforming existing methods.
• The proposed frequency-guided and context aggregation strategies are effective in handling complex segmentation tasks.
• The framework shows great potential for clinical applications requiring high-accuracy segmentation of medical images.