AI to diagnose invisible brain abnormalities in children with epilepsy

(Press-News.org) Scientists have developed an AI-powered tool that detects 64% of brain abnormalities linked to epilepsy that human radiologists miss. 

MELD Graph is an AI tool that could drastically change the care for 30,000 patients in the UK and 4 million worldwide with one cause of epilepsy, researchers say.

The study, published today in JAMA Neurology by a team at King’s College London and University College London (UCL), shows how the tool significantly improves the detection of focal cortical dysplasia’s (FCDs) which is a leading cause of epilepsy.

Researchers say the tool will speed up diagnosis times, get patients the surgical treatment they need quicker, and reduce costs to the NHS by up to £55,000 per patient.

In the UK, 1 in 100 people are affected by epilepsy. 1 in 5 people with epilepsy have seizures caused by a structural abnormality (“lesion”) in the brain. FCDs are a common structural cause of epilepsy and in people with this type of epilepsy, seizures are usually not able to be controlled with medications. Surgery to remove the lesion can be an effective and safe way to stop the seizures. However, the challenge is that FCDs can be subtle and difficult to see with the human eye and up to half of these lesions are missed by radiologists. Delays to diagnosis and surgery mean more seizures, more visits to A&E, and more disruption to school, work and home life. 

In the study, the researchers pooled MRI data from 1185 participants – including 703 people with FCD and 482 controls - from 23 epilepsy centres around the world in the Multicentre Epilepsy Lesion Detection project (MELD). Half of the dataset is from children. They then trained the artificial intelligence tool, MELD Graph, on the scans to detect these subtle brain abnormalities that might otherwise go undetected.

Project lead-author, Dr Konrad Wagstyl, from King’s College London, said: “Radiologists are currently inundated with images they have to review. Using an AI-powered tool like MELD Graph can support them with their decisions, making the NHS more efficient, speeding time to treatment for patients and relieving them of unnecessary and costly tests and procedures.”

Co-author Dr Luca Palma, from Bambino Gesù Children's Hospital, Italy, said: “MELD Graph identified a subtle lesion missed by many radiologists in a 12-year-old boy who had daily seizures and had tried nine anti-seizure medications with no improvement to his condition. This tool could identify patients with surgically operable epilepsy and help with surgical planning – reducing risks, saving money, improving outcomes.”

While the tool is not yet clinically available, the research team have released the AI-tool as an open-source software. They are running workshops to train clinicians and researchers around the world, including Great Ormond Street Hospital and the Cleveland Clinic, in how to use it.

First author, Dr Mathilde Ripart from UCL, said “One of the highlights for me is hearing from doctors around the world, including the UK, Chile, India and France have been able to use our tools to help their own patients.”

Co-author Professor Helen Cross, Prince of Wales's Chair of Childhood Epilepsy, President of the International League Against Epilepsy, Consultant Epileptologist at Great Ormond Street Hospital, and Director of the UCL Great Ormond Street Institute of Child Health, OBE said: “Many of the children I see have experienced years of seizures and investigations before we find a lesion. The epilepsy community is searching for ways to speed up diagnosis and treatment. Initiatives such as MELD have the potential to rapidly identify abnormalities that can be removed and potentially cure the epilepsy.”

Co-lead Dr Sophie Adler from UCL said: “This type of research is only possible with international collaboration. We were privileged to work with 75 researchers and clinicians towards this common goal of “no missed epilepsy lesions worldwide””.

ENDS

Ripart M et al., 2025. Detection of epileptogenic focal cortical dysplasia using graph neural networks: a MELD study. JAMA Neurol, Feb 24. doi:10.1001/jamaneurol.2024.5406

END