Utilization of ChatGPT as a reliable aide for differential diagnosis of histopathology in head and neck surgery

Abstract

The rise of artificial intelligence offers promising advancements in diagnostic workflows in healthcare. In oral and maxillofacial surgery, timely and accurate histopathological diagnosis is crucial for effective treatment planning. This study examines Chat Generative Pretrained Transformer (ChatGPT, OpenAI Inc., California) as an aid to providers in generating differential diagnoses for four common maxillofacial pathologies: ameloblastoma, squamous cell carcinoma, mucoepidermoid carcinoma, and pleomorphic adenoma. A retrospective study was conducted with 200 de-identified histopathological cases, evenly divided across the four diagnostic categories. Each case included clinical summaries and histopathological images, which were input into ChatGPT to generate four differential diagnoses. The study evaluated the inclusion and ranking of the correct diagnosis in the differential list using a chi-square goodness-of-fit test. ChatGPT included the correct diagnosis in all cases (100%), ranking it first in 49.5%, second in 32.5%, third in 14.5%, and fourth in 3.5%. Statistical analysis confirmed a significant preference for higher ranking of correct diagnoses (p< 0.001). ChatGPT shows strong reliability in generating accurate differential diagnoses for maxillofacial histopathology, ranking the correct diagnosis in the top two positions in 82% of cases. These results highlight AI’s potential to augment diagnostic workflows and enhance efficiency. • The article aimed to assess the reliability of ChatGPT, a widely available AI platform, to aide in differential diagnosis of common oral and maxillofacial pathologies • A retrospective study was conducted with 200 de-identified histopathological cases, evenly divided across the four diagnostic categories of ameloblastoma, squamous cell carcinoma, mucoepidermoid carcinoma, and pleomorphic adenoma with a clinical summary detailing patient demographics, lesion characteristics, and relevant clinical history. • ChatGPT was provided with the clinical summaries for each case along with digital pathology imaging and was asked to generate a list of four differential diagnoses. • ChatGPT successfully included the correct diagnosis in 100% of cases with prioritizing the correct diagnosis in the top two positions in 82% of cases.

Ähnliche Arbeiten