HMP-MUNet: A Hybrid Deep Learning Framework for Automated Skin Lesion Segmentation in Dermoscopic Images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Deep learning for cutaneous lesion diagnosis faces challenges in accuracy and computational efficiency.
• Existing models struggle with complex morphological variations and require substantial resources.

Purpose of the Study:

• To introduce High-order Multi-scale Parallel Vision Mamba U-Net (HMP-MUNet), a novel framework for efficient and accurate skin lesion segmentation.
• To address limitations of current deep learning approaches in capturing multi-scale features and computational load.

Main Methods:

• Developed a hybrid U-Net integrating a high-order vision state-space module for global context.
• Employed a multi-scale dilated attention fusion network for hierarchical feature extraction.
• Utilized a parallel multi-depth flexible network for computational optimization and reduced parameter count.

Main Results:

• Achieved high segmentation accuracy with a Dice Similarity Coefficient of 95.85% on PH2 and 90.44% on ISIC2018 datasets.
• Model uses only 7.61M parameters, a 72.2% reduction compared to existing Vision Mamba architectures.
• Demonstrated superior accuracy with significantly reduced computational resources.

Conclusions:

• HMP-MUNet offers an efficient and accurate automated solution for skin lesion segmentation, improving diagnostic consistency.
• The lightweight design shows potential for deployment in diverse clinical settings, enhancing skin cancer screening protocols.
• This framework advances computer-aided diagnosis by integrating high-order modeling with practical deployment considerations for broader healthcare accessibility.