ChatGPT in Pediatric Neuro-oncology: Imaging Model-Derived Differential Diagnoses from Hyperintense Brain Tumors on T2 Weighted MRI

Introduction: The introduction of imaging modality capabilities to ChatGPT represents a groundbreaking stride in the field of Artificial Intelligence (AI). Despite extensive research on ChatGPT’s applications in medicine, there remains a lacuna in the literature regarding the ability of this model to discern pediatric brain tumor (PBT) abnormalities on an MRI. This study aims to fill this gap by assessing the potential of the model to identify abnormalities on MRI scans and generate a comprehensive differential diagnosis.

Methods: MRI images of 31 PBTs, and 10 MRI images of normal pediatric neuroimaging were collected from Radiopaedia, and published in literature. Images were restricted to Axial T2 MRI images. ChatGPT was first probed on whether an abnormality exists in the image, and subsequently asked to generate differential diagnoses based on any abnormal findings.

Results: The model recognized abnormalities with a 74.19% success rate on brain tumor radiology and demonstrated 100% accuracy in recognizing normal images. When evaluating its capacity to identify brain tumors within its differentials, the model included an accurate differential diagnosis on 70.97% of cases. Brain tumors were listed as the first differential in 41.94% of cases. This accuracy rose to 58.06% when considering the top two differentials and to 67.74% within the top three. Amongst all differential diagnoses listed, 59.12% were deemed nonsensical for the MRI finding.

Conclusion : This investigation showcases ChatGPT's adeptness at differentiating between abnormal and normal pediatric brain MRI scans. The model's ability to include the correct diagnosis in its differential —identifying brain tumors 71% of the time—is particularly noteworthy. These findings underscore ChatGPT's potential as an emerging tool in the assessment of pediatric brain tumors, indicating that further refinement and development are required to maximize its application in this field.