AI Tool Enhances Speed and Accuracy of Autism, ADHD Diagnoses

A novel artificial intelligence (AI) tool utilizing motion-tracking data has been developed to diagnose autism and attention-deficit/hyperactivity disorder (ADHD) with increased speed and precision. The research, published on July 8, 2025, in *Scientific Reports*, highlights the potential of this tool to transform the diagnostic process for neurodivergent conditions, offering a more efficient alternative to traditional methods that often rely on subjective behavioral assessments.

Led by Dr. Jorge V. José, a professor of physics and adjunct professor of anatomy, cell biology, and physiology at Indiana University, the study indicates that deep learning models trained on high-resolution kinematic data can effectively distinguish between neurotypical and neurodivergent individuals. Dr. José noted that this technology could address the significant delays—up to 18 months—in receiving formal diagnoses, particularly in regions like Indiana.

Participants in the study were equipped with wireless motion sensors while performing touchscreen-based reaching tasks. These sensors captured linear acceleration, angular velocity, and roll-pitch-yaw (RPY) orientation at millisecond intervals. The resulting data was analyzed using a long short-term memory deep learning model, which achieved an overall accuracy of 71.48% in classifying participants into four categories: neurotypical, autism, ADHD, or comorbid autism and ADHD.

The classification accuracy varied based on the type of data used, with RPY data alone yielding the highest accuracy of 67.83%. However, the model showed limitations in accurately diagnosing comorbid cases, reflecting ongoing challenges in clinical settings. Dr. José emphasized the importance of expanding the dataset for future training to improve diagnostic reliability.

In addition to enhancing diagnostic speed and accuracy, the study introduced unique biomarkers, the Fano Factor and Shannon Entropy, which quantify movement severity. Participants displaying more severe symptoms of autism or ADHD exhibited higher entropy levels, suggesting a correlation between movement variability and symptom severity. Dr. José explained that these metrics could aid in assessing the severity of conditions, which is currently challenging for clinicians.

The researchers envision the AI tool as a potential screening mechanism in various settings, including primary care, schools, and telehealth, particularly in underserved areas. The simplicity of the data collection process—estimated to take only 15 minutes—makes it feasible for early intervention strategies. “This technology could allow for more personalized treatment approaches,” stated Dr. José, highlighting its ability to adjust care based on the severity of a patient’s condition.

As the technology continues to evolve, the integration of increasingly reliable and affordable motion-sensing devices, such as those found in smartphones and smartwatches, could further enhance the applicability of kinematic data analysis in diagnosing neurodivergent conditions. This advancement not only promises to streamline the diagnostic process but also to improve outcomes for children with autism and ADHD by facilitating earlier and more accurate interventions.