Researchers at Ben Gurion University are deploying ML to predict epileptic seizures

About 50 million people worldwide suffer from a debilitating neurological disorder called epilepsy, The chances of self-harm due to a sudden attack are always high. Currently, treatment and medicine can provide many patients with a seizure-free life, but quite a few still find it difficult to lead a normal life when the next seizure is imminent. is. However, while drugs can be delayed, technology is being adopted to create a symbiosis of innovation that improves quality of life.

Researcher from Ben Gurion University We have developed a truly unique device with the goal of regaining a normal sensation in millions of patients around the world. Researchers have used machine learning algorithms to create a wearable device called Epiness that accurately predicts and warns of the next epileptic seizure. The device will be further developed and commercialized through the Israeli startup NeuroHelp. This is part of BGU’s Oazis Accelerator ProgramWas founded by BGN technology And Dr. Oren Shriki, who heads the Cognitive Brain Science Department at BGU.

Machine learns, then machine predicts

Epilepsy is a very prevalent and sometimes debilitating neurological disorder. Up to 30% of patients do not respond appropriately to antiepileptic drugs and are always afraid of imminent seizures. For such patients, a viable seizure predictor may provide a substantial improvement in quality of life and make it possible to avoid seizure-related injuries. Current seizure alarms can detect seizures in real time, but cannot provide advance warning of an imminent seizure. Epines, on the other hand, can detect and warn of pending seizures up to an hour before the seizures occur. This allows the patient to be ready for a seizure, a sense of control and safety, as well as a higher quality of life.

“Epileptic seizures put epileptic patients at a variety of preventable risks, including falls, burns, and other injuries,” said Dr. Oren Schriki. “Unfortunately, there is currently no seizure predictor that can alert patients and prepare for the next seizure. Therefore, the machine learning algorithms we have developed can accurately predict an imminent seizure up to an hour ago. The device under development is accurate and user-friendly because the algorithm has shown that the number of EEG electrodes required can be significantly reduced. Prototypes currently being evaluated in subsequent clinical trials. Is developing. Year. “

Epiness is a seizure prediction and detection device based on a new breakthrough combination of EEG-based monitoring of EEG activity and a unique machine learning algorithm. This device combines a wearable EEG device with state-of-the-art software that minimizes the number of EEG electrodes required and optimizes electrode placement on the scalp. Advanced machine learning algorithms filter noise not related to brain activity, extract useful measurements of underlying brain dynamics, and expect no pre-epileptic brain activity and seizures. It is designed to distinguish brain activity.

The new algorithm was developed and tested using EEG data from a large dataset of epilepsy patients monitored for several days prior to surgery. Patient data was divided into either pre-seizure (pre-seizure) or inter-seizure short segments.

“An accurate and easy-to-use seizure predictor is a very necessary and unmet medical need, as epilepsy that is not properly controlled by drug therapy is widespread and reaches up to 30% of epilepsy cases,” he said. Hadar Ron said. NeuroHelp.. “Current seizure alarms can detect seizures during the onset of seizures. Most seizures rely on changes in movement such as muscle spasms and falls. Epines predicts future seizures and patients And its caretaker are unique in that they can have seizures. Precautions and prevention of injuries. This is also the only device based on brain activity rather than muscle movement or heart rate. Epilepsy is drug resistant. I am confident that it will be a valuable tool in the management of seizures. “

Several machine learning algorithms of varying complexity are trained with pre-assigned training data (including 80% of the initial EEG data), with their predictive and electrode-dependent performance being the remaining 20% ​​of the data. It was evaluated by. The algorithm with the best predictive performance reached 97% level of accuracy and maintained near-optimal performance (95%) even with relatively few electrodes.

Josh Peleg, CEO BGN technology, BGU’s technology transfer companyNeuroHelp, Spin-off BGN technology, Recently established as part of BGU Oazis AcceleratorFormed by Yazamut360 BGU Entrepreneurship Center. Further develop and commercialize innovative solutions for the benefit of people suffering from epilepsy. Earlier this month NeuroHelp Based on a unique combination of brain research and artificial intelligence know-how developed in Dr. Shriki’s lab, he won the highest award at the Silicon Negev Startup Competition, a key recognition of the outstanding potential of this technology. “