Large language model-assisted radiology reporting in a single-radiologist implementation: a retrospective cohort...

Area of Science:

• Medical Imaging Informatics
• Artificial Intelligence in Healthcare
• Radiology Workflow Optimization

Background:

• Radiologist burnout is prevalent, impacting approximately 40% of US radiologists.
• Large language models (LLMs) show potential for improving workflow efficiency in radiology.
• Limited real-world data exists on the implementation and impact of LLMs in clinical radiology settings.

Purpose of the Study:

• To evaluate the impact of an LLM-assisted workflow on radiologist efficiency.
• To assess radiologist satisfaction and adoption drivers using the Unified Theory of Acceptance and Use of Technology (UTAUT) framework.

Main Methods:

• Retrospective cohort study at Mayo Clinic Arizona involving a single fellowship-trained abdominal radiologist.
• Comparison of baseline and post-implementation periods using a custom generative pre-trained transformer (based on ChatGPT Enterprise Model 5.2).
• Analysis of inter-study interval time as a proxy for interpretation time, with assessment of UTAUT constructs.

Main Results:

• LLM assistance significantly reduced inter-study intervals for outpatient CT interpretation (10-11.5 min reduction, p<0.01).
• No significant efficiency improvement was observed for MRI interpretation (p>0.05).
• The radiologist reported improved work-life balance for CT but neutral satisfaction for complex MRI templates; training required 10 hours.

Conclusions:

• LLM-assisted workflows can enhance efficiency for standardized CT studies, potentially reducing documentation burden.
• No clear efficiency benefits were observed for MRI in this sample, highlighting the importance of task-technology fit.
• Performance expectancy and task-technology fit are key drivers for LLM adoption in radiology, necessitating alignment with task complexity.