(0955) Picture-Perfect Diagnosis: How ChatGPT Is Transforming Otologic Health Using Otoscopic Photos

Introduction: ChatGPT is a powerful artificial intelligence model that is swiftly integrating into the healthcare domain. This study assesses ChatGPT's efficacy in analyzing otoscopic images of various otologic diseases, aiming to evaluate its potential as a diagnostic tool for ear diseases and its capacity to potentially transform the realm of otology.

Methods: We created an image database containing the following otologic diseases: otitis media, hemotympanum, cerumen impaction, cholesteatoma, middle ear effusion, otomycosis, glomus tympanicum, osteoma, tympanic membrane perforation, and tympanostomy tube presence. For each condition, we used two otoscopic photos, resulting in a total of 20 photos for evaluation by ChatGPT (version 4.0). We prompted ChatGPT to assess each image with the question, 'What is wrong with this ear?' This process was repeated for 5 consecutive days. A board-certified neuro-otologist evaluated the responses from ChatGPT for correctness (on a scale of 0-3) and consistency (on a scale of 0-3). Additionally, the neuro-otologist assessed each case for complexity, categorizing them as low, moderate, or high.

Results: Among the 20 otoscopic photos, 17 were rated as having low complexity, 2 as moderate complexity, and 1 as high complexity. The overall mean correctness score for ChatGPT was 1.51 (SD = 0.94), with 81% of the responses rated as at least partially correct. The overall mean consistency score was 2.05 (SD = 0.944). The interclass correlation coefficient for the stability of ChatGPT's correctness was 0.589. We did not find a statistically significant difference in the average correctness and consistency scores of ChatGPT relative to case complexity.

Conclusions: ChatGPT-4, though in its nascent phase, has shown substantial potential in diagnosing various otologic disorders through otoscopic image analysis. Further research and development are essential to solidify and expand upon these promising findings.