Topic: ESOPRS 2021 ePoster sessions
Time: Sep 17, 2021 16:00 Amsterdam, Berlin, Rome, Stockholm, Vienna, 15:00 London
(plain text version here)
Assessing the diagnostic power and management recommendations of artificial intelligence on orbital fracture in computed tomography scans
Author: Pelin Celiker
ePoster Number: 126,00
Purpose
In this study, we evaluate the performance of ChatGPT4 (GPT) in the analysis of facial CT for diagnosis of orbital fractures along with its recommended management, compared with the assessments and recommendations of oculofacial plastic surgeons.
Methods
51 cases of facial CT images following head trauma were obtained from open-source online image search, and cases including two or more views (axial and coronal) were included. CT images included in the study were as follows: 2 without orbital fractures, 9 with orbital roof fractures, 12 with medial wall fractures, 20 with orbital floor fractures and 8 with multiple orbital wall fractures.
Each case was assessed for identification of fractures, laterality, and treatment recommendations including medical and surgical management. GPT was given a prompt to assess the CT images along with making recommendations. Separately, an oculofacial plastic surgeon blinded to the image collection was asked to assess the images and to make recommendations purely on the CT images. Performance of CT analysis by GPT was compared to surgeons with the radiologist’s interpretation serving as the gold-standard.
Results
GPT and the surgeon correctly identified the presence of an orbital fracture in 76.47% and 100% cases respectively (p=0.99). The fractured wall was identified correctly in 100% of cases by the surgeon, whereas in 39.21% of cases by GPT (p=0.01). GPT accurately identified the laterality of fracture in 37.25% of cases, which was lower than the surgeon at 100% though statistically insignificant (p=0.74). Lastly, recommendation of surgical management was accurately predicted 49.02% of the time by GPT which was lower than 100% by surgeons though statistically insignificant (p=0.99).
Conclusion
GPT can identify the presence of an orbital fracture on CT images in a majority of cases, but has significant deficiencies compared to oculofacial trauma surgeons in its ability to identify the fractured wall. A larger study could be conducted to further elucidate the capabilities of artificial intelligence in interpreting clinical imaging and assisting in clinical decision making.
Additional Authors
| First name | Last name | Base Hospital / Institution |
|---|---|---|
| Diane | Wang | West Virginia University |
| Omar | Sadat | West Virginia University |
| Kareem | Ibrahim-Bacha | West Virginia University |
| Richard | Cui | West Virginia University |
| Eric | Moran | West Virginia University |
| Taylor | Potter | West Virginia University |
| John | Nguyen | West Virginia University |
Abstract ID: 25-372
