Epilepsy is a devastating disorder affecting an estimated 50-150 individuals per 100,000 people globally, with most incidence hypothesized to burden the low-income countries. While progress has been made in recent years to develop novel drug based therapeutics, roughly one third of cases is considered drug resistant, with no significant improvement rendered to the patients after attempted treatment with 2-3 additional lines of medication.
In the last 20-30 years an exciting novel treatment, involving surgically targeting and excising the zone in the brain responsible for the genesis of epilepsy, has emerged. The epilepsy surgery, originally believed to be applicable in ~5% cases, has been refined continuously with recent studies indicating that up to ~50% of cases in certain epilepsy subtypes could be amenable to surgical intervention.
Surgical targeting of epilepsy is a complex, multi-step procedure requiring tight coordination between healthcare providers and clinical neuroscientists.
The problem that remains with the surgical treatment is that it is invasive and requires a significant amount of careful planning and brain mapping. In the first stage of the procedure, the patient may be operated on to have a set of tiny electrodes installed on the surface of the brain. These electrodes record brain activity and are programmed to detect any abnormal changes of activity that can indicate seizures. Since there are many electrodes working together, it is possible to estimate the area responsible for the generation of the seizure based on the overlapping signals from the mapping electrodes. Moreover, the neuroscientists are able to gently stimulate the electrodes one by one, invoking different motor functions of the patient, such as speech, movement, object recognition, to identify regions that are healthy and responsible for the normal functioning. Once data are collected, in the second stage of treatment the surgeons are able to target the zones that contribute to epilepsy and are not involved in normal function by returning the patient to the operating room for a meticulous brain surgery and repair of the surrounding tissue.
Artificial intelligence methods have been recently developed to help guide the above process and make it less laborious and time consuming. With more efficient pipelines of data analysis, machine learning algorithms are trained on massive EEG data sets, in some cases comprising thousands of patient cases, to map the brain and identify the focal point of epilepsy. Invasive readings are often coupled with non-invasive methods, such as surface EEG headcaps, MRI and PET-scanning to train the algorithm to recognize the region of the epilepsy onset solely based on non-invasive means in future patients.
For example, at the University College London (UK) the Multicentre Epilepsy Lesion Detection (MELD) project developed an algorithm using MRI scans from over 1,000 patients across 22 epilepsy centers to identify abnormalities in drug-resistant focal cortical dysplasia (FCD). FCDs are abnormal brain regions causing epilepsy, often undetectable using standard MRIs. Researchers analyzed brain features from MRI scans and trained the algorithm with labeled examples to help guide surgeons in approaching the brain. The MELD project used the largest MRI cohort of FCDs to date, applicable to patients over three years old with suspected FCD.
While AI development and training may be resource consuming, it is a necessary step towards creating non-invasive and accessible method for mapping epilepsy, making treatment more patient focused and easy to deliver. Additionally, it may enable online diagnosis and consultation of the patients that live in the parts of the world where sophisticated invasive assessment methods, beyond conventional EEG, are not available.
Photo by Anna Shvets
