Utilizing Large Language Models for the Analysis of Video Data in Early Attention Deficit Hyperactivity Disorder Detection in Children
Abstract
This study proposes a novel framework leveraging large language models (LLMs) for early detection of attention deficit hyperactivity disorder (ADHD) in children based on video data analysis. The pipeline begins with behavioral sessions where video recordings are collected and stored systematically. These videos undergo preprocessing involving frame extraction, image processing, and feature extraction to identify key behavioral patterns. The extracted features are analyzed using LLMs to gain insights into attention patterns, movements, and behavioral trends. LLMs are particularly suited for this task as they can process multimodal data and contextualize subtle behavioral cues within a broader diagnostic framework, offering a robust tool for nuanced analysis. Subsequently, an ADHD detection model is trained to classify the likelihood of ADHD, providing high- and low-risk predictions. Results are visualized through a dashboard for clinicians or researchers, enabling further refinement of the detection process. This framework aims to support early ADHD diagnosis with data-driven insights, potentially improving the intervention strategies.