MAPSeg: self-supervised colorectal polyp segmentation via memory-augmented framework and synthetic polyp simulation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Colorectal cancer often develops from polyps, necessitating early detection via colonoscopy.
• Supervised automatic polyp segmentation shows promise but requires extensive annotated datasets, limiting clinical use.

Purpose of the Study:

• To introduce MAPSeg (Memory-Augmented Polyp Segmentation), a fully self-supervised framework for annotation-free colorectal polyp segmentation.
• To address the limitations of supervised methods by developing an unsupervised approach.

Main Methods:

• MAPSeg employs an anomaly detection paradigm trained on healthy colon images.
• It integrates SIMPO (Simulation of Polyps) for realistic synthetic polyp generation.
• A memory-augmented encoder models normal tissue structures.