Computer-Aided Classification of Gastrointestinal Lesions in Regular Colonoscopy

Area of Science:

• Gastroenterology
• Artificial Intelligence
• Medical Imaging

Background:

• Gastrointestinal lesion diagnosis relies on expert interpretation of colonoscopic videos.
• Current methods like chromoendoscopy can be time-consuming.
• Accurate classification of lesions, especially serrated adenomas, is crucial for colorectal cancer prevention.

Purpose of the Study:

• To evaluate computer-based diagnosis of gastrointestinal lesions from colonoscopic videos.
• To develop a novel technique for tissue classification and lesion severity discrimination.
• To compare the AI system's performance against clinical expertise and competitor methods.

Main Methods:

• Developed a framework combining machine learning and computer vision for virtual biopsy.
• Utilized regular (white light and narrow band) colonoscopic videos.
• Created a database of colonoscopic videos with expert-annotated ground truth and histology.

Main Results:

• The AI system demonstrated capability in diagnosing gastrointestinal lesions.
• The technique can potentially reduce the need for time-consuming procedures like chromoendoscopy.
• The study investigated the impact of 3D shape features on classification accuracy.

Conclusions:

• Computer-aided diagnosis shows promise in gastroenterology.
• This technique may enhance efficiency and accuracy in identifying and classifying gastrointestinal lesions.
• Further research can refine AI tools for real-time clinical decision support in colonoscopy.