Sri Harsha Boppana, MBBS, MD1, Manaswitha Thota, MD2, Gautam Maddineni, MD3, Sachin Sravan Kumar Komati, 4, C. David Mintz, MD, PhD5 1Nassau University Medical Center, East Meadow, NY; 2Virginia Commonwealth University, Richmond, VA; 3Florida State University, Cape Coral, FL; 4Florida International University, Florida, FL; 5Johns Hopkins University School of Medicine, Baltimore, MD Introduction: Colorectal cancer is a leading cause of cancer-related deaths, highlighting the need for early polyp detection and malignancy risk assessment. This study presents an AI-driven model using the ERCPMP-v5 dataset to improve real-time identification, classification, and risk assessment of colorectal polyps, enhancing clinical decision-making and patient outcomes. Methods: The ERCPMP-v5 dataset includes 796 high-resolution images and 21 annotated endoscopic videos from 217 patients. The images were standardized for size and quality. A Vision Transformer (ViT) model was used to extract deep image features, which were combined with demographic data (age, gender), morphological classifications (Paris, Kudo Pit, JNET), and pathological diagnoses (e.g., Tubular, Villous, Adenocarcinoma). Three Random Forest classifiers were trained: one for predicting morphological characteristics based on Paris classifications, another for pathology categorization (e.g., Hyperplastic, Adenocarcinoma), and a third for malignancy risk stratification, assigning polyps to risk levels (low, moderate, high). Model performance was evaluated using an 80-20 training-test split, with accuracy, precision, recall, F1-score, and AUC-ROC metrics. Statistical analysis examined correlations between polyp characteristics and patient demographics, providing insights into tailored screening and intervention strategies. Results: The AI model demonstrated excellent performance across all tasks. The morphology classifier achieved an accuracy of 94.94% (precision: 0.93, recall: 0.95, F1-score: 0.94), accurately distinguishing Paris classifications such as 0-IIa and 0-IIb. The pathology classifier reached 99.05% accuracy, successfully identifying polyp types, including adenocarcinoma, tubular, and villous. The malignancy risk classifier achieved 99.58% accuracy (AUC-ROC: 1.0, recall: 96%, F1-score: 0.98), accurately categorizing polyps into risk categories: 23% high risk, 35% moderate risk, and 42% low risk. Demographic analysis revealed that hyperplastic and tubular polyps were most common in individuals aged 60–70, whereas adenocarcinoma was more prevalent in patients over 70. Males exhibited larger polyps and a higher incidence of adenocarcinoma than females. Discussion: This AI-driven model excels in real-time detection, classification, and malignancy risk assessment of colorectal polyps, enhancing clinical decision-making and patient outcomes. Incorporating demographic factors enables personalized screening strategies.
Disclosures: Sri Harsha Boppana indicated no relevant financial relationships. Manaswitha Thota indicated no relevant financial relationships. Gautam Maddineni indicated no relevant financial relationships. Sachin Sravan Kumar Komati indicated no relevant financial relationships. C. David Mintz indicated no relevant financial relationships.
Sri Harsha Boppana, MBBS, MD1, Manaswitha Thota, MD2, Gautam Maddineni, MD3, Sachin Sravan Kumar Komati, 4, C. David Mintz, MD, PhD5. P4776 - Revolutionizing Colorectal Cancer Detection: AI-Powered Real-Time Classification and Risk Stratification of Polyps, ACG 2025 Annual Scientific Meeting Abstracts. Phoenix, AZ: American College of Gastroenterology.
