DBCM-net:dual backbone cascaded multi-convolutional segmentation network for medical image segmentation

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Medical image segmentation is crucial for clinical applications but faces challenges with low-contrast and blurred boundaries in endoscopic/dermoscopic images.
• Existing single and dual-encoder architectures struggle with detail preservation and context integration, leading to segmentation inaccuracies.

Purpose of the Study:

• To develop an advanced segmentation network, the Dual Backbone Cascaded Multi-Convolutional Segmentation Network (DBCM-Net), to address limitations in medical image segmentation.
• To improve the accuracy and detail preservation in segmenting challenging medical images, particularly those from endoscopic and dermoscopic sources.

Main Methods:

• Employed a cascaded dual-encoder architecture featuring a Multi-Axis Vision Transformer and a Vision Mamba encoder for multi-scale feature extraction.
• Introduced Global and Local Fusion Attention Block (GLFAB) and Depthwise Separable Convolution Attention Module (DSCAM) for enhanced feature representation and integration.
• Utilized a Feature Refinement Fusion Block (FRFB) for refining feature maps within the cascaded structure.

Main Results:

• Achieved high Dice coefficients: 94.93% on CVC-ClinicDB, 91.93% on ISIC2018, and 92.73% on ACDC.
• Demonstrated superior performance compared to state-of-the-art methods across multiple medical image segmentation datasets.
• Showcased effective preservation of fine-grained edge details in segmented images.

Conclusions:

• DBCM-Net offers a robust solution for precise medical image segmentation, outperforming existing methods.
• The proposed architecture effectively integrates multi-scale features and attention mechanisms for improved segmentation accuracy.
• The DBCM-Net shows significant potential for clinical applications requiring high-fidelity medical image segmentation.