Assessment of ChatGPT Analysis of Common Neuroradiological Spinal Abnormalities

Introduction: Before referral to a spinal surgeon, it is not uncommon for patients with back pain to have obtained imaging. With increasing use of electronic medical records, patients typically have access to these images prior to discussion with their physicians. With the growing popularity of ChatGPT and the development of its image analysis software, the authors hoped to assess the ability of ChatGPT to interpret radiological images of the spine.

Methods: The authors provided ChatGPT with the images of common spinal abnormalities and assessed ChatGPT’s interpretation of the image as well as its diagnosis. The authors further assessed if ChatGPT identified the correct image modality, and whether errors were anatomical or related to image density. Statistical analysis was performed with Python (Python Software Foundation, Wilmington, DE).

Results: The authors queried ChatGPT with 11 radiographic images of the spine. ChatGPT identified the image modality correctly 11 of 11 times, demonstrating the ability to differentiate between Computerized tomography (CT) and Magnetic Resonance Imaging (MRI). ChatGPT also demonstrated an understanding of the differences between T1 and T2 MRI sequences. However, ChatGPT only was able to correctly analyze only 64% of images. Frequent errors were observed in regards to misinterpretation of density (18%) or anatomic structures (18%). The diagnosis provided by ChatGPT was only found to be accurate in 5 of 11 responses (45%).
ChatGPT did not provide any diagnoses that had a worse prognosis than the actual diagnosis, but it did provide a more benign diagnosis than the actual diagnosis in 27% of images.

Conclusion : ChatGPT demonstrates an excellent understanding of the modalities of radiographic imaging of the spine. However, it is limited in its ability to analyze radiographic images with errors stemming from anatomical inaccuracies and difficulty interpreting the density of images. Despite these limitations it is still able to accurately diagnose radiographic abnormalities nearly half of the time. While ChatGPT demonstrates an elementary understanding of radiographic images of the spine, patients should be cautioned in using the software to interpret their images.