An adaptive paradigm for computer-aided detection of colonic polyps

Area of Science:

• Medical Imaging
• Computational Pathology
• Gastroenterology

Background:

• Current computer-aided detection (CADe) for colonic polyps often assumes uniform local environments.
• This assumption fails in the colon due to variations like haustral folds, leading to missed detections, especially for small polyps.

Purpose of the Study:

• To develop an adaptive CADe paradigm for colonic polyps that accounts for varying local environments.
• To improve the detection and characterization of small colonic polyps located on haustral folds.

Main Methods:

• Decomposed colon structure into flat walls and haustral folds.
• Developed adaptive local environment descriptions tailored to each sub-structure.
• Focused on local geometrical measures for initial polyp candidate (IPC) detection and false positive (FP) reduction.

Main Results:

• Successfully detected small polyps (<10 mm) on haustral folds that were previously missed.
• Reduced false positives by retaining small polyps on folds during the FP reduction phase.
• Demonstrated the significant impact of adaptive local environment descriptions.

Conclusions:

• The adaptive paradigm enhances CADe for colonic polyps, particularly small ones on folds.
• This approach is expected to improve polyp detection, volume measurement, and management planning.
• Adaptively describing local environments is crucial for accurate colonic polyp analysis.