A - 179 AI-Driven Diagnostics for Child Neuropsychology: Promising but Insufficient.

Abstract

This research tested the capacity of a current generative AI model (i.e., ChatGPT-4) to accurately diagnose neuropsychological issues in children. The secondary aim was to explore the AI's potential in assisting in cognitive functioning evaluation, with the goal of improving the accessibility, efficiency, and accuracy of diagnosis. The study had three phases. First, the diagnostic information of 15 children (including the results of a standard neuropsychological battery and basic demographic information) was provided to the generative AI model to suggest a diagnosis. These results were then compared with the child's confirmed diagnoses. Next, a logistic regression was conducted using a large dataset on the most commonly missed diagnosis by the AI model, which identified the common factors not initially included, but necessary for a more accurate diagnosis by ChatGPT-4. The third phase involved re-analyzing the initial datasets in the same manner as the first, but with the addition of salient information determined by the logistic regression. The results of this study suggest that current generative AI models perform well interpreting scores under sanitized conditions. However, AI also struggles to interpret the effect of context. Although the model performed marginally better with the use of additional scores and information, it still struggled to capture the nuances of context-sensitive conditions. AI technology can assist diagnosis, but due to failures of understanding in-situ symptomatology, the use must remain limited. Further advancements can enhance neuropsychological assessment for more accuracy and accessibility. Clinical implications and future directions are discussed.