An AI-Based Colonic Polyp Classifier for Colorectal Cancer Screening Using Low-Dose Abdominal CT.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Colorectal cancer (CRC) screening relies on accurate detection of polyps.
• Non-invasive methods like Computed Tomography Colonography (CTC) and Virtual Colonoscopy (VC) offer high accuracy.
• Existing methods require robust AI for efficient polyp detection.

Purpose of the Study:

• To develop and evaluate an AI-based framework for automatic polyp detection in virtual colonoscopy (VC).
• To assess the framework's performance on low-dose CT scans and its utility in CRC screening.
• To enhance polyp visualization using a novel Fly-In (FI) approach.

Main Methods:

• An AI framework combining automatic colon segmentation and polyp detection was developed.
• A Fly-In (FI) visualization approach generated 2D projections fused with 3D colon models for synthetic image creation.
• A RetinaNet model was trained on synthetic images for polyp detection, achieving 94% f1-score and 97% sensitivity.
• A simulation platform evaluated the FI approach's performance on low-dose CT scans using an AI restoration algorithm.

Main Results:

• The AI-trained RetinaNet model demonstrated high performance in polyp detection.
• The Fly-In (FI) approach effectively visualized the colon, aiding gastroenterologists in CRC detection.
• The AI restoration algorithm successfully enhanced low-dose CT images for 3D colon reconstruction and visualization.
• Radiologist evaluations showed high relative sensitivities (92% at 30 KV, 99.5% at 60 KV) for the FI approach on low-dose CTC.

Conclusions:

• The proposed AI framework significantly improves polyp detection accuracy in virtual colonoscopy.
• The Fly-In (FI) approach, enhanced by AI, shows great promise for non-invasive colorectal cancer screening, especially with low-dose CT.
• The study highlights the potential of AI in medical imaging for early disease detection and diagnosis.