An artificial intelligence (AI) algorithm to detect subtle brain abnormalities that cause epileptic seizures has been developed. The abnormalities, known as focal cortical dysplasias (FCDs), can often be treated with surgery but are difficult to visualize on an MRI. The new algorithm is expected to give physicians greater confidence in identifying FCDs in patients with epilepsy.
The work, which was part of the Multicentre Epilepsy Lesion Detection (MELD) project, appeared in Brain “Interpretable surface-based detection of focal cortical dysplasias: a Multi-centre Epilepsy Lesion Detection study.” Konrad Wagstyl, PhD, and Sophie Adler, PhD, both from University College London, led an international team of researchers on the work.
To develop the algorithm, the team quantified features of the brain cortex—such as thickness and folding—in more than 1,000 patient MRI scans from 22 epilepsy centers around the world. They then trained the algorithm on examples labeled by expert radiologists as either being healthy or having FCD.
FCD is a malformation of cortical development in the brain, and have a propensity to cause epilepsy that does not respond to medication. Although FCDs are typically treated with surgery, MRI scans often look normal, making diagnosis challenging.
“In clinically ambiguous images, where the need for algorithms is greatest, such insight would enable physicians to determine whether features identified by the classifiers are likely to be lesional in origin,” explained the researchers in their article.
About one percent of the world’s population has epilepsy, a disorder of the brain that is characterized by frequent seizures. Drug treatments are available for the majority of them, but 20–30% do not respond to medications.
Of patients who have epilepsy that have an abnormality in the brain that radiologists do not see on MRI scans, FCD is the most common cause.
“Our algorithm automatically learns to detect lesions from thousands of MRI scans of patients. It can reliably detect lesions of different types, shapes and sizes, and even many of those lesions that were previously missed by radiologists,” said Hannah Spitzer, PhD, a postdoctoral researcher in machine learning at Helmholtz Zentrum Münchenand co-first author on the article.
Overall, their algorithm was able to detect the FCD in 67% of cases in the cohort (538 participants).
Previously, 178 of the participants had been considered MRI negative, which means that radiologists had been unable to find the FCD abnormality. But the MELD algorithm was able to identify the FCD in 63% of these cases. This is important, because if doctors can confidently identify FCDs, then surgery to remove them can provide a cure.
“We put an emphasis on creating an AI algorithm that was interpretable and could help doctors make decisions,” said Mathilde Ripart, research assistant at University College London’s Great Ormond Street Institute of Child Health and the other co-first author on the article. “Showing doctors how the MELD algorithm made its predictions was an essential part of that process.”
