Cutting Edge to Cutting Time: Can ChatGPT Improve the Radiologist's Reporting?

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Informatics

Background:

• Structured reports (SR) in radiology offer advantages over free text (FT) but face implementation challenges.
• Large language models like GPT-4 present an opportunity to enhance radiology reporting workflows.

Purpose of the Study:

• To develop and evaluate a GPT-4 powered web application for converting free text radiology reports to structured reports.
• To assess the impact of this AI tool on reporting time and report quality.

Main Methods:

• Two radiologists reported 30 abdominopelvic CT scans across control (traditional) and AI-assisted sessions.
• A GPT-4 web application was used to structure free text into SR in the AI-assisted session.
• Reporting times (Total Reporting Time and Case Turnaround Time) and quality scores were compared.

Main Results:

• AI-assisted reporting showed a trend towards decreased Total Reporting Time and Case Turnaround Time, though not statistically significant.
• Quality scores of AI-assisted reports were comparable to traditional reports.
• Excluding editing time, AI-assisted Case Turnaround Time was statistically significantly reduced compared to control.

Conclusions:

• A GPT-4 powered tool can effectively generate structured radiology reports from free text.
• The AI tool has the potential to reduce reporting times without negatively impacting report quality.
• Optimizing editing time within the AI-assisted workflow presents an avenue for further efficiency gains.