A recent publication from our lab in the Journal of Magnetic Resonance Imaging highlights how large language models can be used to improve understanding of cardiac MRI image quality before scans are even performed. In Pre-Imaging Clinical Factors Associated With Cardiac MR Image Quality Using Large Language Model-Enabled Data Extraction, co-first authors Hong Yu, MD, PhD, and Masha Bondarenko, BS, together with Ali Nowroozi, MD, Adrian Serapio, BS, and Jae Ho Sohn, MD, MS, demonstrated that routinely available clinical information extracted from electronic health records using an LLM can identify patients at higher risk for poor cardiac MRI image quality. Across 1,006 clinical cardiac MRI examinations, cognitive and communication impairment and respiratory compromise were independently associated with reduced image quality, suggesting opportunities to proactively adjust workflows and preparation before scanning.
This work represents an important example of the lab’s focus on translating AI into clinically actionable imaging infrastructure. By showing that structured variables can be reliably extracted from routine documentation at scale and linked to downstream imaging performance, the study establishes a pathway toward anticipatory quality management in cardiac MRI and more efficient scanner utilization.
