New York – September 30, 2026 -- Researchers at NYU Langone Health have built an artificial intelligence system that tracks the daily case exposure of radiology residents and flags missing pathologies with more than 90 percent accuracy, according to a study published online in Academic Radiology.
AI system pinpoints training gaps that traditional case logs miss
The tool monitors which conditions residents encounter during clinical rotations and automatically identifies underexposed pathologies, then assigns supplemental teaching cases to close the gap. Study author Vinay Prabhu, an associate professor in NYU Grossman School of Medicine's Department of Radiology, said residents seeing 30 cases in a day typically encounter 29 routine exams and only one uncommon condition, leaving gaps in exposure to rarer diseases they will face throughout their careers.
ChatGPT-4o selects three to five targeted cases per resident each night
The research team first built a curriculum of key conditions residents should see in their first three years of training, developed by faculty across five imaging specialties -- abdominal, musculoskeletal, brain, pediatric, and chest imaging -- based on board-certification exam requirements. ChatGPT-4o then reads the summary sections of residents' daily clinical reports and selects three to five prioritized teaching cases each night, which appear on residents' workstations alongside regular clinical assignments for review with supervising physicians.
Approach widens pathology breadth without cutting real patient caseloads
Michael P. Recht, chair of NYU Langone's Department of Radiology, said the system marks a shift from uniform training curricula toward automated, data-driven personalization tied directly to where residents work. Study co-author Matthew G. Young said existing methods -- lecture courses, shared teaching files, textbooks, and question banks -- vary widely between programs and fail to reliably track whether residents see an adequate range of pathologies.
Team plans to expand tracking into four additional imaging subspecialties
The researchers intend to extend the tool to cardiac imaging, nuclear medicine, breast imaging, and interventional radiology. The study was funded in part by a grant from the Committee of Interns & Residents/Service Employees International Union Healthcare Patient Care Trust Fund. Contributing authors included NYU Langone's Antonio Verdone, Erin Alaia, Anna Chen, Luoyao Chen, Sumit Chopra, Ryan Cummings, Jay Karajgikar, Jane Ko, Renata La Rocca Vieira, Shailee Lala, Evan Stein, Naomi Strubel, Danielle Toussie, and William Walter, alongside Malte Westerhoff of Visage Imaging in Berlin, Germany.