Wednesday, December 3 | 8:20 a.m.-8:30 a.m. | W1-SSNPM03-3 | Room S402
In this presentation, attendees will learn how ChatGPT can assist radiologists with proper ICD-10 coding of PET/CT exams.
With radiologists often working up to five extra hours each week just on administrative tasks, large language models are increasingly being looked at for their potential to reduce administrative workloads, according to presenter Tristan Ruhwedel, MD, of Charité - Universitätsmedizin Berlin in Germany.
In their study, Ruhwedel and colleagues compared the accuracy of ICD-10 coding by two experienced nuclear medicine physicians: One used ChatGPT o1 for assistance and the other used only traditional search engines.
On 100 different PET/CT reports, ChatGPT-assisted coding resulted in six incorrect codes, while 18 errors were found in manual coding. Furthermore, median manual coding time was 135 seconds, compared with just eight seconds with the assistance of ChatGPT. In addition, the median time difference between the two coding approaches was 127 seconds.
“Higher accuracy of ICD-10 codes might reduce billing errors and insurance rejections, while saved time could be reallocated to patient care,” the authors wrote. “As total-body PET/CT becomes more widely adopted and scan durations decrease, these efficiencies might prevent up to an hour of overtime daily in a center performing 30 PET/CT scans, potentially helping alleviate administrative burdens and reduce burnout risk.”
The researchers also noted that given “the complexity of PET/CT reports, similar approaches might be feasible for other imaging modalities like diagnostic CT or MRI.”
Explore the potential of LLMs for billing applications at this Wednesday morning talk.
