AI pinpoints radiology residents’ educational gaps, helping expand personalized training
New data detail how artificial intelligence can be harnessed to improve radiology resident education by identifying gaps in clinical exposure and reinforcing concepts missed during training.
Radiology residency programs typically provide residents with broad clinical exposure, but the cases and pathologies encountered can vary considerably based on clinical volume, subspecialty assignments and the patients seen at an institution. Experts recently sought to determine whether AI could help identify these differences and provide personalized educational opportunities to address them.
Published in Academic Radiology, the analysis highlights the potential AI possesses beyond its diagnostic utility. The paper describes the use of a “Precision Education” program that helps increase radiology residents’ exposure to important pathologies across multiple subspecialties without substantially reducing their clinical case interpretation volumes. The approach used a large language model to analyze residents’ clinical cases, identify gaps in exposure and provide supplemental teaching that targeted areas where individual residents had fallen below predefined curriculum targets.
“Current strategies to address pathology exposure gaps typically include lecture-based didactics, faculty-shared teaching files, and self-directed supplemental learning (e.g., textbooks, question banks, videos),” Vinay Prabhu, MD, MS, with NYU Langone Health, Department of Radiology, and colleagues noted. “These methods, however, can be inconsistent, imprecise, and variable between trainees, faculty members and programs. Furthermore, they often fail to mimic routine clinical scenarios. Therefore, there is a need for structured educational tools to precisely address pathology exposure gaps within the clinical learning environment.”
The study used curriculum outlining the important pathologies residents should expect to encounter during postgraduate years two through four. Daily clinical reports from residents were analyzed using ChatGPT-4o prompts to identify the important pathologies seen during routine clinical work. Residents were then provided with curated teaching cases, with priority given to pathologies for which their clinical exposure remained below defined targets.
Researchers compared pathology exposure and clinical case volumes before and after implementing the program to determine whether targeted education could improve resident know-how. ChatGPT-4o demonstrated greater than 91% precision and recall in identifying the important pathologies encountered by residents, suggesting that the model could reliably track pathology exposure from clinical reports. This led to residents encountering significantly more unique pathologies in every radiology subspecialty evaluated.
In abdominal imaging, median unique pathology exposure increased from ranges of 75–107 before the intervention to 93.5–144 afterward; musculoskeletal imaging increased from 43.5–70 to 73–99, while neuroimaging increased from 32.5–38 to 64.5–79. Pediatric imaging exposure increased as well, from 39.5–49 and unique pathologies to 82–96.5, while thoracic imaging increased from 42.5–56 to 49.3–85.3. All increases were considered statistically significant.
Importantly, the intervention did not come at the expense of residents’ clinical training. Median numbers of live cases interpreted did not significantly decrease after implementation, with the exception of postgraduate year 3 abdominal imaging cases. The findings suggest that AI-generated educational recommendations could be a beneficial supplement to the clinical cases residents encounter during routine training.
Read more here.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.