Award: ACG Outstanding Research Award in the Biliary/Pancreas Category
Ji Young Bang, MD, MPH1, Adrian Saftoiu, MD2, Jayapal Ramesh, MD1, Anca Udristoiu, PhD3, Lucian Gruionu, PhD3, Elena Codruta Gheorghe, MD2, Gabriel Gruionu, PhD4, Udaykumar Navaneethan, MD1, Robert Hawes, MD1, Shyam Varadarajulu, MD1 1Orlando Health, Digestive Health Institute, Orlando, FL; 2Universitatea de Medicina si Farmacie Carol Davila din Bucuresti, Bucharest, Bucuresti, Romania; 3Universitatea din Craiova, Craiova, Dolj, Romania; 4Indiana University School of Medicine, Indianapolis, IN
Introduction: As outcomes in pancreatic cancer are stage dependent, high technical proficiency in EUS is required for early detection and accurate diagnosis of pancreatic lesions. We evaluated performance of artificial intelligence enhanced EUS (AI-EUS) to detect solid and cystic lesions in the pancreas.
Methods: An AI-EUS software was developed using convolutional neural network (CNN) model that included 32,713 linear-array EUS frames (training and testing phases) of normal, solid and >10mm cystic pancreatic lesions from 202 patients. A prospective, nonrandomized, comparative study was conducted in patients referred for pancreatic cancer screening and suspected solid or cystic lesions in the pancreas. EUS findings were examined concurrently in real-time by two independent expert examiners, one using conventional EUS and another with AI-EUS. EUS procedures were projected to separate monitors with examiners blinded to alternate assessments. Main outcome was detection of solid pancreatic mass lesions. Secondary outcome was detection of pancreatic cysts.
Results: 269 patients were included (mean age 66.2 years [SD 14.7]; male 45.4%) between January to June 2024. There was 97.8% (95% CI, 92.4-99.7) concordance between conventional EUS and AI-EUS (p=0.497) for detection of solid masses (n=92; mean size 31.4 mm [SD=15.6]) that were biopsy proven to be adenocarcinoma in 67, metastatic lesion 2, neuroendocrine tumor 9, lymphoma 1, solid pseudopapillary neoplasm 1, GIST 1, and benign disease in 11. For detection of normal pancreas (n=73), there was 100% (95% CI, 95.1-100) concordance between conventional EUS and AI-EUS. For pancreatic cysts (n=104; mean size 30.2 mm [SD 25.2]), as compared to AI, conventional EUS diagnosed more cysts (100% [95% CI 96.5-100] vs. 83.7% [95% CI 75.1-90.2], p< 0.001), particularly those < 15 mm. (Figure).
Discussion: In this first human trial, the concordance between conventional EUS and AI-EUS was greater than 95% for detection of solid pancreatic mass lesions. AI has potential to standardize technical performance at EUS, which can further optimize clinical outcomes in pancreatic cancer. Additional refinements to CNN model incorporating Doppler techniques could improve detection of small pancreatic cysts.
Figure: Figure Legend: A pancreatic mass measuring 14 by 12 mm on conventional EUS (A), identified on AI-EUS as a red area (B). A pancreatic cyst measuring 12 by 9 mm on conventional EUS (C), identified on AI EUS as a blue area (D).
Disclosures:
Ji Young Bang: Boston Scientific Corporation – Consultant. Olympus America Inc. – Consultant.
Adrian Saftoiu: Medical Softverse – Stock-privately held company.
Jayapal Ramesh indicated no relevant financial relationships.
Anca Udristoiu: Medical Softverse – Stock-privately held company.
Lucian Gruionu: Medical Softverse – Stock-privately held company.
Elena Codruta Gheorghe: Medical Softverse – Stock-privately held company.
Gabriel Gruionu: Medical Softverse – Stock-privately held company.
Robert Hawes: Fujifilm – Consultant. GIE Medical – Consultant. Olympus America Inc. – Consultant.
Shyam Varadarajulu: Boston Scientific Corporation – Consultant. Medtronic – Consultant. Olympus America Inc. – Consultant.
Ji Young Bang, MD, MPH1, Adrian Saftoiu, MD2, Jayapal Ramesh, MD1, Anca Udristoiu, PhD3, Lucian Gruionu, PhD3, Elena Codruta Gheorghe, MD2, Gabriel Gruionu, PhD4, Udaykumar Navaneethan, MD1, Robert Hawes, MD1, Shyam Varadarajulu, MD1, 50, Performance of a Novel Artificial Intelligence System for Detection of Solid Pancreatic Mass Lesions at Real-Time Endoscopic Ultrasound (EUS), ACG 2024 Annual Scientific Meeting Abstracts. Philadelphia, PA: American College of Gastroenterology.
