The potential of DeepSeek for AI-aided diagnosis of antibody-positive autoimmune encephalitis: a single-center, retrospective, observational study.

Background: Autoimmune encephalitis (AIE) is challenging to diagnose, especially in primary hospitals in China with limited medical resources. DeepSeek, a newly developed AI, shows potential as a cost-effective tool for improving diagnostic efficiency. However, no studies have evaluated the diagnostic accuracy of DeepSeek for AIE. Methods: This retrospective study included 100 patients with anti-neuronal antibody-positive AIE treated at Ruijin Hospital, Shanghai Jiao Tong University School of Medicine. After removing personally identifiable information, antibody results, and history of immunotherapy from patients' medical histories, the following information was sequentially input into DeepSeek: sex, age, chief complaint, medical history, EEG findings, head MRI description, and cerebrospinal fluid (CSF) results. The positive rates of AIE diagnoses predicted by DeepSeek were then categorized as most likely diagnosis, differential diagnosis, and total diagnosis. Results: Using DeepSeek, the probabilities of AIE appearing as the most likely diagnosis and total diagnosis accuracy were 49 and 65%. When patient data were input stepwise, both the total diagnosis accuracy and the most likely diagnosis accuracy did not significantly increase. AIE patients with anti-MOG and anti-GABAbR positivity had predicted total diagnostic positivity rates of 88 and 100%, respectively. Patients presenting with headache and epilepsy were more likely to be diagnosed with AIE (96 and 100%). Conclusion: DeepSeek shows limited positive diagnostic accuracy for predicting the diagnosis of AIE. The application of this new AI technology could be used to promote early screening for AIE in primary hospitals in China, improve medical education, and lead to research advances in AIE.