An Efficient Multi-Task Synergetic Network for Polyp Segmentation and Classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Colonoscopy is crucial for early detection and resection of polyps, preventing colorectal cancer.
• Accurate segmentation and classification of polyps from colonoscopic images are vital for diagnosis and treatment planning.

Purpose of the Study:

• To propose an efficient multi-task synergetic network (EMTS-Net) for concurrent polyp segmentation and classification.
• To introduce a polyp classification benchmark for exploring task correlations.
• To enhance the accuracy and efficiency of polyp detection in colonoscopic images.

Main Methods:

• Developed an EMTS-Net framework comprising an enhanced multi-scale network (EMS-Net) for coarse segmentation, EMTS-Net (Class) for classification, and EMTS-Net (Seg) for fine segmentation.
• Utilized coarse segmentation masks to aid precise polyp localization and classification.
• Implemented a random multi-scale (RMS) training strategy and offline dynamic class activation mapping (OFLD CAM) to optimize network performance.

Main Results:

• Achieved an average mDice of 0.864 for polyp segmentation.
• Obtained an average AUC of 0.913 and an accuracy of 0.924 for polyp classification.
• Demonstrated superior performance, efficiency, and generalization compared to state-of-the-art methods.

Conclusions:

• The proposed EMTS-Net effectively performs concurrent polyp segmentation and classification.
• The novel framework significantly improves the accuracy of polyp detection in colonoscopic images.
• EMTS-Net represents a promising advancement for colorectal cancer prevention through enhanced endoscopic analysis.