AI model offers accurate and explainable insights to support autism assessment

A deep learning model was shown to achieve up to 98% accuracy in distinguishing autistic from neurotypical participants
Human brain activity with plexus lines.. External cerebral connections in the frontal lobe. Communication, psychology, artificial intelligence or AI, neuronal informations or cognition concepts illustration with copy space. - stock illustration

AI model offers accurate and explainable insights to support autism assessment

A deep learning model was shown to achieve up to 98% accuracy in distinguishing autistic from neurotypical participants
Scientists have developed and tested a deep-learning model that could support clinicians by providing accurate results and clear, explainable insights ÂÌñ»»ÆÞ“ including a model-estimated probability score for autism.
The model, outlined in a study published in eClinicalMedicine (a journal from The Lancet), was used to analyse resting-state fMRI data ÂÌñ»»ÆÞ“ a non-invasive method that indirectly reflects brain activity via blood-oxygenation changes.
In doing so, the model achieved up to 98% cross-validated accuracy for Autism Spectrum Disorder (ASD) and neurotypical classification and produced clear, explainable maps of the brain regions most influential to its decisions.
ASD diagnoses have increased substantially over the past two decades, partly reflecting greater awareness, expanded screening, and changes to diagnostic criteria and clinical practice. Early identification and access to evidence-based support can improve developmental and adaptive outcomes and may enhance quality of life, though effects vary.
However, because the current diagnosis primarily relies on in-person and behavioural assessments ÂÌñ»»ÆÞ“ and the wait for a confirmed diagnosis can stretch from many months to several years ÂÌñ»»ÆÞ“ there is an urgent need to improve assessment pathways.
The researchers hope that, with further validation, their model could benefit autistic people and the clinicians who assess and support them by providing accurate, explainable insights to inform decisions.
The study was the result of a final-year undergraduate project by BSc (Hons) Computer Science student Suryansh Vidya, supervised by Dr Amir Aly, and researchers from the School of Engineering, Computing and Mathematics at the ÂÌñ»»ÆÞ. They were in turn supported by researchers from the UniversityÂÌñ»»ÆÞ™s School of Psychology and the CIDER ÂÌñ»»ÆÞ“ Cornwall Intellectual Disability Equitable Research, part of the Peninsula Medical School. 

There are more than 700,000 autistic people in the UK, and many others are waiting to be assessed.

Because diagnosis still depends on a specialist, in-person behavioural evaluation, the journey to a confirmed decision can take many months ÂÌñ»»ÆÞ“ and, in some areas, years. Our work shows how AI can help: not to replace clinicians, but to support them with accurate results and clear, explainable insights, including a model-estimated probability score, to help prioritise assessments and tailor support once further validated.

Amir AlyDr Amir Aly
Lecturer in Artificial Intelligence and Robotics

Using the Autism Brain Imaging Data Exchange (ABIDE) cohort, which included 884 participants aged seven to 64 across 17 sites, the team processed the data with established pipelines and ran a side-by-side comparison of explainability methods. Gradient-based techniques performed best, and the resulting maps were broadly consistent across these pipelines, highlighting the brain regions most influential to the modelÂÌñ»»ÆÞ™s decisions.
The research is already being taken forward by PhD researcher Kush Gupta, a co-author on the current study, incorporating different kinds of multimodal data and machine learning models with the objective of developing a robust and generalisable AI-driven model that could support clinicians in autism assessment all over the world. This complements Dr AlyÂÌñ»»ÆÞ™s broader research programme, including the use of robots to support autistic people, and developing AI methods for analysing health-sector data.

We have shown that artificial intelligence has the potential to act as a catalyst for early autism detection and advancing diagnostic accuracy. However, some of Robert FrostÂÌñ»»ÆÞ™s words come to mind ÂÌñ»»ÆÞ“ ÂÌñ»»ÆÞ˜the woods are lovely, dark and deep, but we have miles to go before we sleepÂÌñ»»ÆÞ™. In the same way, these are early prototypes which require further validation and research.

Rohit ShankarProfessor Rohit Shankar
Professor in Neuropsychiatry and Director of the CIDER group

 

A project developed with students and early career researchers

Suryansh Vidya graduated with a First from the BSc (Hons) Computer Science (Artificial Intelligence) course in 2024

I was initially drawn to this course, and to the ÂÌñ»»ÆÞ, by the blend of theoretical computer science foundations and extensive practical work. I loved that it wasn't just academic theory but also real-world application.
The opportunity to spend a year in industry during the course was particularly appealing, and it has proved transformative. It's what secured me an Amazon internship and eventually a full-time role with the company, where IÂÌñ»»ÆÞ™m currently working as a software development engineer. I was also attracted to the strong research culture, and being able to engage in research with Dr Amir Aly and his team ÂÌñ»»ÆÞ“ while also gaining industry experience ÂÌñ»»ÆÞ“ was invaluable.
I chose to focus on autism diagnosis for my final year project because of the significant clinical need. There are over 200,000 people waiting for NHS assessments, with diagnosis timelines stretching anywhere between two and eight years. Dr Aly's lab was already conducting research in this area, making it an ideal opportunity to contribute to meaningful work with real clinical potential.
The challenge was developing AI that clinicians could trust, which meant creating models that weren't just accurate but could help them explain their decisions. What I found most rewarding was the systematic benchmarking of interpretability methods, and discovering that gradient-based approaches like Integrated Gradients worked best for brain connectivity data, contrary to previous findings in image analysis.
BSc (Hons) Computer Science (Artificial Intelligence) graduate Suryansh Vidya
The biggest surprise was consistently identifying visual processing regions (particularly the calcarine sulcus and cuneus) as critical biomarkers across all preprocessing pipelines. When we validated our findings against independent genetic studies, the results confirmed that we'd identified genuine neurobiological markers rather than dataset artifacts.
This convergence of computational and biological evidence was incredibly satisfying, and I hope the findings of our study will make a real difference for clinicians and those with autism in the future.
 

Kush Gupta is a PhD researcher exploring the use of artificial intelligence in the diagnosis of Autism Spectrum Disorder (ASD)

My interest in artificial intelligence (AI) began during my four years at Medtronic, a healthcare-based multinational company. I worked on applying AI techniques to address real-world healthcare challenges, and it sparked a strong curiosity about the potential of AI to transform healthcare delivery. To deepen my expertise, I pursued a Masters in Image Processing and Computer Vision at the University of Bordeaux, which exposed me to state-of-the-art AI methods being adopted across the healthcare sector.
With the number of autism cases rising significantly over the past decade, I chose to embark on my PhD as I wanted to contribute meaningful research that could improve early diagnosis and intervention. I was drawn to the ÂÌñ»»ÆÞ in particular for the opportunity to join Dr Amir AlyÂÌñ»»ÆÞ™s research group developing AI technology for healthcare. I was especially inspired by the emphasis on bridging the gap between academic research and everyday clinical use, which mirrors my own professional vision.
Our recent paper in eClinicalMedicine explores the use of AI to support and enhance ASD diagnosis. Rather than clinicians relying solely on traditional diagnostic tools based on clinical observations, the proposed AI model can assist them in making more accurate and timely diagnostic decisions. Importantly, we incorporated explainable AI methods, enabling us to identify critical brain regions often associated with ASD, which could support clinicians in planning tailored interventions.
PhD researcher Kush Gupta
The study highlighted that the visual and auditory processing regions of the brain are particularly affected in individuals with autism. These findings could explain the common challenges in social interaction, attention deficits, limited eye contact, and delayed language development.
Looking ahead, I plan to extend this research by incorporating multiple modalities and diverse datasets, considering variations across race, gender, and ethnicity. The aim is to develop a robust and generalisable AI-driven diagnostic model that can reliably support clinicians worldwide.
 

Gain skills in real-world applications, AI theory, and creating tools for Industry 4.0, while exploring its ethical impact and engaging with top research:

Pepper robot
 

Read more information linked to this article

More related news

A picture from Getty Images depicting a depressed woman sitting on the couch at home
Engineering Getty

School of Engineering, Computing and Mathematics ÂÌñ»»ÆÞ“ our research

Carnewas and Bedruthan Steps in Cornwall at sunset

CIDER ÂÌñ»»ÆÞ“ Cornwall Intellectual Disability Equitable Research

Signal transmitting neuron or nerve cell in the brain

Centre for Robotics and Neural Systems (CRNS)