Automated Generation of Narrative Sleep Reports Utilizing Portable Electroencephalogram Data Through ChatGPT

Sleep is a very important activity, but many people do not know their own sleep conditions. A sleep test personalizes sleep quality assessment and detects potential sleep disorders by measuring biological signals. The rise in sleep-related issues has necessitated the development of automated testing methods. Machine learning plays a pivotal role in interpreting sleep data and determining sleep stages. However, the generation of detailed reports and tailored recommendations still demands expert intervention. Automating the report generation to provide personalized sleep insights is a crucial and desired step for the future of sleep healthcare. Recently emerged Generative AI, such as ChatGPT, has attracted considerable attention in recent years. It can generate new sentences and images from input data. In this study, we investigate the practicality and applicability of using ChatGPT to generate narrative sleep reports for sleep test. In our proposed method, GPT-4 receives the information about the sleep habits of the participants and the sleep assessment automatically summarized by the rule-based algorithm. In the evaluation, we used in-home sleep EEG data obtained from 100 subjects by S'UIMIN inc. The generated reports were evaluated by experienced technicians and physicians on a 5- point scale for medical correctness and appropriateness as informative reports. The results of the evaluation showed that 60 % of the reports were the acceptable or above range in both aspects. While more than half of the results were judged to be above the acceptable range, differences between the generative AI and humans were also identified. Whereas humans comment on semantically weighted important findings such as medication and subjective insomnia, ChatGPT tends to make broad, shallow and flat comments on the input data. These facts suggest that although practical report generation only using generative AI is at present not easy, generative AI is a promising tool for improving the efficiency of physicians and technicians work.