AI algorithm finds ‘needle in a haystack’ brain abnormalities on MRI

Some patients with epilepsy that is not controlled with medications have brain abnormalities that are missed on MRI scans. Some of these patients have an invasive procedure involving probes being inserted into their brain to find the abnormality.

The new study co-authored by Konrad Wagstyl and Sophie Adler from the Computational Anatomy Team at the WCHN and GOSICH reveals an AI algorithm that can find some of these hidden MRI abnormalities.

Lead author Konrad Wagstyl said: “Epilepsy surgery is a procedure to remove an area of the brain where seizures originate. In children with a single seizure location, like in this study, surgery can offer seizure freedom in approximately 70 per cent. However, in some patients the brain abnormality is not found on MRI scans and the patients require probes to be inserted to help find it. We wanted to see if we could use artificial intelligence (AI) to help find these subtle brain abnormalities on MRI.”

In this study researchers used deep learning to train an algorithm to detect these brain abnormalities on MRI. The algorithm was then tested on patients with severe epilepsy who despite extensive investigations, the location of the abnormality was not clear.

The algorithm successfully detected 86 per cent of the focal cortical dysplasias, a type of isolated brain abnormality known to cause epilepsy.

Epilepsy is not just one condition, but a group of many different epilepsies with one thing in common: a tendency to have seizures that start in the brain. Anyone can develop epilepsy at any time of life and happens in people of all ages, races, and social classes but is most commonly diagnosed in children and in people over 65.

Around one in 100 people are diagnosed with epilepsy in the UK. The patients in this study had a type of epilepsy caused by a brain abnormality present from birth.

The researchers are now running a clinical trial where they aim to help neurosurgeons plan brain implantations with the algorithm.

They are also running an international collaborative study, the Multi-centre Epilepsy Lesion Detection project (MELD) with epilepsy centres from around the world, training and testing the algorithm on patients from the collaborating hospitals.

Links

• Article in Epilepsia 
• Wellcome Centre Human Neuroimaging 
• Great Ormond Street Institute for Child Health
• MELD project