Machine learning has been used to create an algorithm which can diagnose epilepsy more consistently, which could allow doctors to perform life-changing surgery to remove or even cure epilepsy in many patients.
Epilepsy is a long-term brain condition which can lead to repeated seizures in affected individuals. A new artificial intelligence (AI) tool developed by an international team led by researchers at University College London (UCL) in the UK has been shown to better detect subtle abnormalities in the brain which are a leading cause of epilepsy.
The algorithm automatically learns to detect lesions from thousands of MRI scans of patients and has been shown to reliably detect lesions of different types, shapes and sizes, and many lesions that were previously missed by radiologists.
Focal cortical dysplasia (FCD) are areas of the brain that have developed abnormally and are a leading cause of drug-resistant epilepsy. FCDs are typically treated with surgery but identifying the abnormalities in the first place can be a real challenge.
Often a life-long condition with serious repercussions, early indication and treatment for epilepsy would be a great relief for millions of people and their families around the world.
About one percent of the world’s population have the serious neurological condition characterised by frequent seizures. This number is consistent in Australia according to the Epilepsy Foundation. In the UK, where the AI tool was developed, some 600,000 people are affected. While drugs treatments are available for most, medications are ineffective in 20-30% of people with epilepsy.
In children who have had surgery to control their epilepsy, FCD is the most common cause, and in adults it is the third most common cause. FCD is also the most common cause of brain abnormalities which have gone unnoticed in magnetic resonance imaging (MRI) scans but have caused epilepsy in people.
The Multicentre Epilepsy Lesion Detection project (MELD) used more than 1000 patient MRI scans from 22 global epilepsy centres to develop an algorithm to find FCDs. The study is the largest on MRI scans of FCDs to date, meaning it is able to detect all types of FCD, and can be run on the MRI scans of any patient over three years old who may have an FCD.
Many MRI scans hosting FCDs can look normal, making their identification very difficult for clinicians. So, the AI tool was developed to aid in their search.
In developing the algorithm, the researchers defined cortical features (details in the outer layer of the brain’s cerebrum) from the MRI scans of 300,000 locations across the brain. They trained the AI to look at features like how thick or folded the cortex/brain surface was.
To train the algorithm, the team exposed it to examples labelled by expert radiologists as being either healthy or having FCD. The algorithm detected FCDs in 67% of cases out of the 538 participants.
Out of the cohort, 178 participants were considered by the radiologists to be MRI negative – meaning the radiologists did not find any FCD. Yet the MELD algorithm found the FCD in 63% of these cases.
Greater ability to detect the abnormalities will allow doctors to perform life-changing surgery to remove it and cure epilepsy in many patients.
Unfortunately the authors write that “false positives were common in both patients and controls.” Not only could a false positive lead to unnecessary surgery which would be very expensive, it could potentially be dangerous. The researchers are working to minimise this.
Co-first author Mathilde Ripart from UCL’s Great Ormond Street Institute of Child Health says: “We put an emphasis on creating an AI algorithm that was interpretable and could help doctors make decisions. Showing doctors how the MELD algorithm made its predictions was an essential part of that process.”
“This algorithm could help to find more of these hidden lesions in children and adults with epilepsy and enable more patients with epilepsy to be considered for brain surgery that could cure the epilepsy and improve their cognitive development.”
“We hope that this technology will help to identify epilepsy-causing abnormalities that are currently being missed,” adds co-senior author Dr Sophie Adler, also from the Great Ormond Street Institute of Child Health. “Ultimately it could enable more people with epilepsy to have potentially curative brain surgery.”