Diagnostic Performance of ChatGPT-4.0 in Histopathological Analysis of Gliomas: A Single Institution Experience.

Abstract

This study aimed to evaluate the performance of ChatGPT-4.0 as a diagnostic support tool for pathologists in identifying different types of gliomas based on histopathological data and to compare its performance with that of another artificial intelligence tool (Gemini 2.5 Pro). A retrospective analysis was performed on 25 cases with histopathological descriptions. The dataset, anonymized for patient confidentiality, included clinical details such as age, sex, and site, along with two histological images for each case, obtained from the archive files of the Anatomic Pathology section, Department of Medical, Surgical Sciences and Advanced Technologies "G.F. Ingrassia" University of Catania, Italy. ChatGPT-4.0 was tasked with generating diagnoses, which were classified as correct, similar, or different when compared to the pathologists' conclusions and the diagnoses provided by Gemini. ChatGPT-4.0 achieved a diagnostic accuracy of 88%, correctly identifying 22 out of 25 cases. No significant differences in diagnostic performance were observed between male and female patients. The AI performed exceptionally well in diagnosing glioblastomas, with a 100% accuracy rate, while two oligodendrogliomas and one astrocytoma IDH-mutant G3 were misdiagnosed. A comparative evaluation with Gemini 2.5 Pro was also conducted, although its contribution was limited to a qualitative comparison based on the same dataset. ChatGPT-4.0 demonstrated moderate accuracy in the histopathological diagnosis of gliomas, with little variability depending on glioma subtype. While its performance highlights potential for future integration into clinical workflows, significant improvements are required to ensure its reliability and effectiveness in diagnostic applications. Trial Registration: ce 165/2015/PO.