MRI-reading AI homes in on the brain abnormalities that cause drug-resistant epilepsy

Epilepsy still has no cure, and while many cases can be managed with anti-seizure medications, as many as 30% of cases are drug-resistant and therefore require more intensive treatment methods.

One such approach, resective surgery, in which the area of the brain that’s causing a patient’s seizures is removed, can be very successful. But these procedures are extremely limited, not only by the high risk associated with delving into many areas of the brain but also by how difficult it can be for physicians to pinpoint an exact cause of epilepsy.

Aiming to make that task quite a bit simpler is a group of researchers from University College London that published a study in the journal Brain this month describing an artificial intelligence algorithm they built to detect the source of seizures in the brain.

The researchers focused specifically on cases of focal cortical dysplasia. FCDs—regions where brain cells have developed and congregated abnormally, affecting the formation of the cortex—have been cited as a leading cause of drug-resistant epilepsy and can be treated with surgery but are often tricky to spot in standard MRI scans.

In the study, the research team collected more than 1,000 brain MRIs from 22 epilepsy centers around the world. They built an AI algorithm that was trained to quantify each scan, looking at about 300,000 locations on the brain to measure thickness and detect folds in the cortex, the outermost layer of the brain, and comparing those to scans that expert radiologists had determined to have FCDs.

Once trained using half the scans, the algorithm was put to the test on the remaining MRIs, representing nearly 550 patients. The AI achieved a sensitivity of just under 60% in spotting FCDs in the scans, a level that went up to 67% when a border zone was added around the lesions in the scans.

Additionally, the algorithm was able to accurately detect FCDs in 63% of a group of 178 patients whose brain MRIs had previously been declared healthy and FCD-free by radiologists.

The AI tool can be used on any patient who is at least three years old, has undergone an MRI and is suspected of having an FCD.

The researchers said that the algorithm could potentially boost the number of drug-resistant epilepsy cases successfully treated with surgery by detecting more cortical abnormalities than can be spotted by humans alone. They noted that FCDs are the most common cause of surgeries to control epilepsy in children and the third most common in adult patients.

“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,” Konrad Wagstyl, Ph.D., co-senior author of the study, said in a press release. “Roughly 440 children per year could benefit from epilepsy surgery in England.”