Innovative Large Language Model Enhances Patient Understanding of Radiology Reports

In a significant advancement for medical communication, Stanford University researchers have developed a groundbreaking large language model named RadGPT aimed at helping patients better comprehend their radiology test results. Often, medical reports like MRI and CT scan interpretations contain complex terminology that can be difficult for individuals without medical training to understand. For example, a diagnosis such as "mild intrasubstance degeneration of the posterior horn of the medial meniscus" may not be meaningful to most patients.

RadGPT addresses this challenge by extracting key concepts from radiologist reports and translating them into clear, patient-friendly explanations. This enables individuals to grasp what their imaging results mean and to formulate relevant questions for their healthcare providers. The model can also suggest follow-up questions to facilitate more engaging and informed discussions.

The research, published in the Journal of the American College of Radiology, involved developers using 30 sample radiology reports to train the system by extracting five core concepts from each report. Human radiologists reviewed the explanations to ensure safety and accuracy, confirming that RadGPT produces reliable information without hallucinations or harmful content.

While RadGPT currently depends on radiologists creating reports, the model's potential to empower patients is substantial, especially since US patients have had legal access to their electronic health records, including radiology reports, since 2021. Dr. Curtis Langlotz, Stanford professor and senior author, emphasizes that this technology may not only improve patient understanding but also enhance communication between healthcare professionals and patients.

Despite its promising capabilities, RadGPT is still in early stages and requires further validation in clinical settings. It operates by interpreting structured radiology reports rather than raw scan images and is designed as a supplementary educational tool rather than a diagnostic system.

The aim is to reduce cognitive overload for radiologists, who often experience high levels of mental workload, and to foster more active patient engagement in their health. With ongoing advancements, RadGPT and similar AI models could transform the way medical information is communicated, making healthcare more accessible and patient-centered.