Attraction Propagation: A User-Friendly Interactive Approach for Polyp Segmentation in Colonoscopy Images

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Computational Geometry

Background:

• Accurate segmentation of colorectal polyps is crucial for early cancer detection during optical colonoscopy (OC).
• Existing interactive segmentation methods often require multiple user inputs, increasing procedural complexity.
• Integration into computer-aided diagnosis (CAD) workflows necessitates efficient and accurate pre-processing tools.

Purpose of the Study:

• To introduce and evaluate Attraction Propagation (AP) as a novel, user-friendly interactive segmentation technique for colorectal polyps.
• To assess the performance of AP compared to existing state-of-the-art methods in terms of accuracy and computational speed.
• To demonstrate the compatibility of AP with the pre-processing stage of computer-aided diagnosis systems used in optical colonoscopy.

Main Methods:

• Developed an interactive segmentation approach based on Attraction Propagation (AP).
• Utilized a single foreground seed for polyp segmentation, simplifying user interaction.
• Integrated AP into a systematic procedure compatible with optical colonoscopy (OC) and computer-aided diagnosis (CAD) pre-processing.
• Conducted experiments on a diverse dataset comprising 1691 OC images.

Main Results:

• Attraction Propagation (AP) demonstrated superior performance compared to state-of-the-art methods.
• AP achieved higher accuracy in segmenting colorectal polyps.
• AP exhibited significantly faster calculating speed, enhancing efficiency.
• The method proved effective across a challenging and distinct dataset.

Conclusions:

• Attraction Propagation (AP) provides an effective and efficient solution for interactive colorectal polyp segmentation.
• The user-friendly, single-seed approach of AP simplifies the segmentation process for optical colonoscopy.
• AP's performance advantages in accuracy and speed make it a valuable tool for computer-aided diagnosis systems.