CGMA-Net: Cross-Level Guidance and Multi-Scale Aggregation Network for Polyp Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Colorectal cancer (CRC) poses significant mortality and morbidity risks.
• Colonoscopy is crucial for CRC prevention and control.
• Manual polyp segmentation is time-consuming and requires expert input.

Purpose of the Study:

• To develop an automated polyp segmentation method for colonoscopy images.
• To address limitations of existing methods, such as high similarity between polyps and mucosa.
• To improve the accuracy and efficiency of polyp detection in colonoscopy.

Main Methods:

• Proposed Cross-level Guidance and Multi-scale Aggregation Network (CGMA-Net).
• Incorporated Cross-level Feature Guidance (CFG) for region highlighting.
• Utilized Multi-scale Aggregation Decoder (MAD) for feature dependency capture.
• Employed Details Refinement (DR) with asynchronous convolution and attention for enhanced detail and global information.

Main Results:

• CGMA-Net achieved state-of-the-art performance on benchmark datasets (Kvasir-SEG and CVC-ClinicDB).
• Achieved Dice Similarity Coefficient (DSC) of 91.85% on Kvasir-SEG and 95.73% on CVC-ClinicDB.
• Demonstrated strong generalization capabilities with DSC scores of 86.25% and 86.97% on respective datasets.
• CGMA-Net achieved these results with relatively fewer parameters.

Conclusions:

• CGMA-Net significantly advances automated polyp segmentation in colonoscopy.
• The proposed method offers a more efficient and accurate alternative to manual segmentation.
• This technology has the potential to improve early detection and management of colorectal cancer.