P4782 - Optimizing Post-Polypectomy Surveillance: A Multimodal Machine Learning Model for Predicting Colorectal Polyp Recurrence and Tailoring Surveillance Intervals

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 (CRC) ranks among the leading causes of cancer‐related morbidity and mortality worldwide, with adenomatous polyps recognized as its principal precursor. Although removal of these lesions during colonoscopy substantially lowers CRC risk, polyp recurrence remains common, underscoring the need for effective post‐polypectomy surveillance. Recent advances in machine learning and multimodal data integration offer promising avenues to refine surveillance strategies and alleviate associated clinical burdens.

Methods: We employed the ERCPMP-v5 dataset, comprising demographic, morphological, and pathological metadata alongside over 430 anonymized colonoscopy images and videos from 217 patients. Continuous variables were scaled and categorical features one-hot encoded. Visual embeddings were extracted from images using a Vision Transformer (ViT). A fusion network combined metadata and image features to (1) predict polyp recurrence as “likely” or “not likely” and (2) recommend personalized surveillance intervals based on risk. Model performance was assessed on a held-out test set using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC).

Results: Our multimodal model achieved 99.89 % accuracy, 100 % precision, 99.53 % recall, and a 99.77 % F1-score. The AUC reached 1.0000, indicating perfect separation of recurrence and non-recurrence cases. Based on prediction confidence, patients were stratified into three surveillance cohorts: low-risk (follow-up every 1–3 years), moderate-risk (6–12 months), and high-risk (3–6 months). These results demonstrate the model’s robust ability to both identify recurrence and suggest evidence-based follow-up schedules.

Discussion: This study confirms the feasibility and efficacy of a multimodal predictive model for post-polypectomy surveillance. With near-perfect accuracy and discrimination, it represents a significant enhancement to clinical decision support. Tailored surveillance recommendations promise to minimize unnecessary procedures for low-risk individuals while ensuring prompt follow-up for higher-risk patients, ultimately improving outcomes and optimizing resource utilization in practice.

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. P4782 - Optimizing Post-Polypectomy Surveillance: A Multimodal Machine Learning Model for Predicting Colorectal Polyp Recurrence and Tailoring Surveillance Intervals, ACG 2025 Annual Scientific Meeting Abstracts. Phoenix, AZ: American College of Gastroenterology.