Dr Bekaii-Saab on AI-Powered Early Detection Tools in Pancreatic Cancer

"For those at risk, certainly at higher risk, this [REDMOD AI] may actually be a game changer... meaning that we're going to see more and more patients being captured earlier in the game, where we actually can make a bigger difference."

Tanios S. Bekaii-Saab, MD, FACP, the David F. and Margaret T. Grohne Professor of Novel Therapeutics for Cancer Research I at Mayo Clinic College of Medicine and Science; the division chair of Hematology/Medical Oncology at Mayo Clinic; the co-leader of the Advanced Clinical and Translational Science Program; and the disease group leader for Gastrointestinal Cancers for the Mayo Clinic Comprehensive Cancer Center, discussed the transformative potential of artificial intelligence (AI) in the early detection of pancreatic cancer.

Bekaii-Saab began by asserting that early detection is the most critical factor for improving outcomes in pancreatic cancer, and several AI tools could aid in this objective. One AI model generating significant buzz in this space is the Radiomics-based Early Detection MODel (REDMOD), developed by Mayo Clinic. This automated, mechanistically grounded model is designed to detect subtle imaging nuances that may be overlooked even by the most experienced radiologists during routine scans.

A study was conducted to develop and validate REDMOD using data and workflows that mirror clinical practice, including CT scans from multiple institutions, imaging systems and protocols. The study utilized nearly 2,000 CT scans from multiple institutions and imaging protocols. Researchers applied the AI model to scans from patients later diagnosed with pancreatic cancer—all of which had been originally interpreted as normal. The results demonstrated that REDMOD identified 73% (AUC 0.82) of prediagnostic cancers at a median of 475 days prior to clinical diagnosis. Notably, the model's sensitivity was nearly twice as nearly twofold higher than that of radiologists overall (38.9%) and almost tripled grew to nearly threefold (68.0% vs 23.0%) when evaluating scans taken more than 24 months before diagnosis.

Beyond its sensitivity, REDMOD showed strong longitudinal stability, with 90% to 92% concordance, and maintained generalizable specificity across multi-institutional and public datasets. Bekaii-Saab emphasized that this technology could be a "game changer" for high-risk populations, as earlier detection significantly increases the opportunity for surgical intervention. Furthermore, capturing the disease earlier allows clinicians to better leverage emerging agents and tailor treatments to the tumor’s biology, Bekaii-Saab emphasized. Ultimately, combining these AI tools with existing clinical workflows will be essential for improving survival in pancreatic cancer, he concluded.