AI Helps Speed Up and Sharpen Autism and ADHD Diagnoses
A team of researchers from Indiana University has developed an innovative AI-powered diagnostic approach that could make it much faster and more accurate to identify autism spectrum disorder (ASD) and attention-deficit/hyperactivity disorder (ADHD). Published in Scientific Reports in July 2025, the study shows how artificial intelligence can analyze minute hand movements to detect neurodivergent conditions — potentially reducing diagnostic times from months to just 15 minutes.
Why This Matters
Getting a formal diagnosis for autism or ADHD can take over a year in some parts of the U.S., such as Indiana, where children often wait up to 18 months for psychiatric assessments. During this waiting period, essential interventions and therapies are delayed — and for many families, that delay can make a huge difference in a child’s development and wellbeing.
Currently, psychiatrists rely on interviews, behavioral assessments, and questionnaires, but there are no biological or quantitative tests to support their evaluations. The symptoms of these disorders vary widely — that’s why they’re often called “spectrum” conditions. One person’s experience of autism or ADHD can look completely different from another’s.
The team at Indiana University wanted to tackle that diagnostic gap. Led by Professor Jorge V. José, a physicist and neuroscientist, and joined by colleagues John I. Nurnberger, Martin Plawecki, and Khoshrav Doctor from the University of Massachusetts Amherst, the group has been working for years to create data-driven tools that bring objectivity to psychiatric diagnosis.
How the AI Test Works
The core idea behind their approach is surprisingly simple. Participants perform a reaching task on a touchscreen: they touch a target that pops up at random intervals, then pull their hand back and repeat the process several times. During the exercise, a sensor attached to the hand captures incredibly detailed motion data — recording acceleration, angular velocity, and orientation (roll, pitch, and yaw) hundreds of times per second.
At first glance, these micromovements are invisible to the human eye. But when processed through advanced deep learning algorithms, they reveal patterns that correlate with different neurodevelopmental profiles. The system examines how sporadic, random, or smooth a person’s movement is. Research has shown that more randomness in motion tends to appear in individuals with autism or ADHD compared to neurotypical individuals.
The AI model uses supervised deep learning, built with LSTM (Long Short-Term Memory) networks and fully connected layers, to categorize participants into one of four groups: autism, ADHD, both autism and ADHD, or neurotypical. Importantly, the algorithm works directly with raw kinematic data rather than just pre-filtered versions, allowing it to pick up subtle fluctuations that might otherwise be lost.
Precision Through Motion Biometrics
The study also introduces a powerful new feature: the ability to measure how severe a disorder might be using biometric indicators. These indicators are based on statistical analyses of motion fluctuations, namely the Fano Factor and Shannon Entropy.
- Fano Factor measures variability — how much a person’s movements deviate from the average pattern.
- Shannon Entropy quantifies randomness or unpredictability in those movements.
Higher values of both suggest greater irregularity, which can indicate more pronounced neurodivergent characteristics. The researchers found that neurodivergent participants (especially those with both autism and ADHD) showed significantly higher variability than neurotypical participants.
To ensure reliability, the team tested how quickly these metrics stabilize. They discovered that the Fano Factor became consistent after about 27 trials, while Shannon Entropy reached stability after around 64 trials. This means the test doesn’t need to be lengthy — in fact, the entire process can be done in roughly 15 minutes.
Impressive Accuracy and Real-World Potential
The AI system achieved a test accuracy of about 71.5% in correctly identifying whether participants had autism, ADHD, both, or neither. That level of accuracy is quite strong for early-stage behavioral diagnostic tools — especially considering that traditional psychiatric evaluations can sometimes show comparable variability between raters.
The researchers believe their system could be used as a screening tool in schools or clinics to prioritize which children should be referred for full diagnostic evaluation. It wouldn’t replace a psychiatrist’s judgment but could act as an early-warning system — ensuring that those who need help the fastest get it sooner.
Graduate researcher Chaundy McKeever described the project as taking “a physicist’s approach to understanding the brain.” The team treats motion as a window into neural processing, analyzing movement irregularities as reflections of differences in brain coordination and control.
From 2018 to Now: How the Research Evolved
This 2025 study builds upon earlier work from 2018, when José and colleagues first discovered that small, unseen movement differences could distinguish autistic participants from neurotypical ones. In that earlier experiment, the group used basic motion sensors and found measurable differences in how participants reached for objects.
Now, with newer Bluetooth-enabled kinematic sensors that capture data at higher resolution — including acceleration, rotation, and spatial orientation — the system can identify even subtler patterns. The addition of AI-driven deep learning dramatically enhanced the analysis, allowing it to spot relationships between data points across hundreds of milliseconds of motion.
A Step Toward Personalized Neurodivergence Care
Beyond diagnosis, this system could help clinicians and therapists track treatment progress. If a child’s movement patterns change after behavioral therapy or medication, the AI could quantify those differences — offering an objective measure of improvement.
For example, patients with severe autism or ADHD may show highly random and inconsistent movements. As therapies help regulate attention and motor control, the AI might detect more consistent motion patterns, helping providers fine-tune treatment intensity and monitor recovery.
That kind of precision could make care more personalized, affordable, and accessible, especially for families with limited resources.
Challenges and Cautions
While the findings are exciting, the researchers acknowledge several limitations. The study’s sample size was relatively small, and not all participants could complete the reaching task due to unrelated motor impairments. Additionally, information about whether participants were taking medication was not available — an important factor, since medications can affect motor control.
The team emphasizes that this technology is meant to complement, not replace, traditional diagnosis. Ethical and practical challenges remain, especially regarding data privacy, sensor accuracy, and overreliance on algorithmic decisions.
Experts also warn that every neurodivergent person is unique — a machine learning model, no matter how advanced, cannot capture the full context of a person’s lived experience. For that reason, integrating human insight remains essential.
Understanding Autism and ADHD
Autism and ADHD are both neurodevelopmental disorders that affect attention, communication, and behavior, though in different ways. Autism typically involves differences in social communication, sensory processing, and repetitive behaviors, while ADHD primarily impacts attention, impulse control, and activity levels.
Both conditions share overlapping features — in fact, many individuals are diagnosed with both autism and ADHD. According to recent estimates, roughly 30–50% of autistic individuals also meet criteria for ADHD. Early diagnosis and intervention can dramatically improve outcomes, which is why any tool that helps identify these conditions faster is so valuable.
Looking Ahead
The Indiana University team envisions future versions of the tool running on smartphones or smartwatches, making large-scale screenings possible even outside of clinics. By turning motion data into a reliable diagnostic signal, their work could transform how neurodivergence is recognized and supported worldwide.
While the technology is still in development, the promise is clear: AI may soon help bridge the gap between waiting lists and timely care, ensuring that children get the support they deserve without months of uncertainty.
Research Reference:
Deep learning diagnosis plus kinematic severity assessments of neurodivergent disorders – Scientific Reports (2025)
