AI-Enabled Assessment of Adult ADHD

adult ADHD, AI-enabled assessment

Prof Marios Adamou OBE and Prof Grigoris Antoniou discuss the problems and solutions of adult ADHD, using an AI-enabled assessment of data

Within the UK, adult ADHD occurrence is increasing, affecting about 3-5% of children and now, 2% of adults.

Attention-deficit hyperactivity disorder (ADHD) is characterised by a persistent and inappropriate pattern of inattention, hyperactivity, and/or impulsivity that causes significant impairment across domains.

Along with these three core symptom domains, people with ADHD also present with deficits in executive functions, behaviour and emotional regulation and motivation.

The problem in the UK is predominantly under-treatment and under-diagnosis

Demand for diagnostic assessments of adults with ADHD has been growing for the last 15 years. This is because there is more awareness amongst the population and health professionals about this condition. Because of this high demand, but also difficulties with recruitment and retention of health professionals across healthcare in the UK, the time people need to wait to receive an ADHD assessment is very long.

We know that this delay to receive a diagnosis and treatment for ADHD is harmful to people. For example, untreated ADHD is linked to more mental health problems, relationship problems, employment problems and substance misuse.

Deploying better technology as a solution to ADHD diagnosis

The bottleneck at the root of the mismatch between demand and supply for diagnosis in adult ADHD is the time of specialist senior doctors and other professionals who are sufficiently trained to diagnose adult ADHD.

The authors have co-developed a new AI technology that identifies clear-cut cases and delivers a proposed diagnostic decision based on patient data while referring the more difficult cases for further assessment by senior clinical specialists. Less reliance on the time of senior clinical specialists will increase the throughput of cases and will reduce waiting lists. (Figures 1 and 2) As a result, deployment of this technology will shorten the time people will need to wait, because a broader range of health professionals will be able to complete the diagnostic assessments quicker.

This is because, using clinical data as input, the AI will be able to guide health professionals about who requires extra assessment by whom, and who doesn’t.

Figure 1: Current assessment practice
Figure 2: Assessment with new AI technology. Note that clear-cut cases are dealt by more junior healthcare professionals, easing the burden on senior clinicians and thus increasing throughput of cases.
Figure 3: Hybrid AI solution

A unique aspect of our solution is that relies on a hybrid AI algorithm that uses a hybrid AI algorithm which combines a machine learning model and a knowledge model: (Figure 3) A machine learning-based model analyses the data that the NHS Trust holds in its computing systems – the same data a clinician would have access to – and produces a diagnostic outcome of ADHD, along with a confidence factor (a measure of confidence). A variety of machine learning algorithms were used, but the best performance was delivered by a decision tree algorithm. Predicted diagnostic performance of this model is around 85%.

In addition, in a number of in-depth interviews with experienced clinicians, knowledge acquisition was used to capture medical knowledge about how the AI system should reach a decision based on the data collected for a particular case. This implicit knowledge was represented as a number of rules (which use the same case data as an input) with three possible outcomes: positive, negative or consult expert.

Clinical and technical validity of AI

The technology was co-developed through a partnership between South-West Yorkshire Partnership NHS Foundation Trust and the University of Huddersfield. After initial development, our team received funding from the NHS AI lab under its inaugural call on AI for Health and Care. This project, which was success- fully finalised in April 2022, demon- strated clinical and technical validity through a trial at an NHS adult ADHD service. The accuracy of our hybrid AI was shown to be around 95% which is an excellent result since the solution is designed to provide decision support, not autonomous decisions.

Further developments in the area of AI for mental health We are now in the process of publishing the validation results. Next steps will be to identify NHS sites for further trials. We are also looking for opportunities of commercialisation, either through a spin-out company or through collaboration with digital healthcare providers.

Finally, we have further developments in the area of AI for mental health, addressing autism and suicide risk assessment.

More information

You can contact us by email at

marios.adamou@swyt.nhs.uk or g.antoniou@hud.ac.uk.
Information can also be found at https://selene.hud.ac.uk/aeaaa/neu- rointel/

An illustrative video is found at:

https://www.youtube.com/watch?v=tygLmWAwBbU

 

Please note: This is a commercial profile

© 2019. This work is licensed under CC-BY-NC-ND.

Contributor Profile

Professor
Centre of AI for Mental Health, University of Huddersfield
Phone: +44 (0)1484 472 147
Website: Visit Website

LEAVE A REPLY

Please enter your comment!
Please enter your name here