Summary: Researchers apply machine learning algorithms to EEG data to create a new method of predicting when epileptic seizures will strike.
Source: University of Sydney.
Epileptic seizures strike with little warning and nearly one third of people living with epilepsy are resistant to treatment that controls these attacks. More than 65 million people worldwide are living with epilepsy.
Now researchers at the University of Sydney have used advanced artificial intelligence and machine learning to develop a generalised method to predict when seizures will strike that will not require surgical implants.
Dr Omid Kavehei from the Faculty of Engineering and IT and the University of Sydney Nano Institute said: “We are on track to develop an affordable, portable and non-surgical device that will give reliable prediction of seizures for people living with treatment-resistant epilepsy.”
In a paper published this month in Neural Networks, Dr Kavehei and his team have proposed a generalised, patient-specific, seizure-prediction method that can alert epilepsy sufferers within 30 minutes of the likelihood of a seizure.
Dr Kavehei said there had been remarkable advances in artificial intelligence as well as micro- and nano-electronics that have allowed the development of such systems.
“Just four years ago, you couldn’t process sophisticated AI through small electronic chips. Now it is completely accessible. In five years, the possibilities will be enormous,” Dr Kavehei said.
The study uses three data sets from Europe and the United States. Using that data, the team has developed a predictive algorithm with sensitivity of up to 81.4 percent and false prediction rate as low as 0.06 an hour.
“While this still leaves some uncertainty, we expect that as our access to seizure data increases, our sensitivity rates will improve,” Dr Kavehei said.
Carol Ireland, chief executive of Epilepsy Action Australia, said: “Living with constant uncertainty significantly contributes to increased anxiety in people with epilepsy and their families, never knowing when the next seizure may occur.
“Even people with well controlled epilepsy have expressed their constant concern, not knowing if or when they will experience a seizure at work, school, travelling or out with friends.
“Any progress toward reliable seizure prediction will significantly impact the quality of life and freedom of choice for people living with epilepsy.”
Dr Kavehei and lead author of the study, Nhan Duy Truong, used deep machine learning and data-mining techniques to develop a dynamic analytical tool that can read a patient’s electroencephalogram, or EEG, data from a wearable cap or other portable device to gather EEG data.
Wearable technology could be attached to an affordable device based on the readily available Raspberry Pi technology that could give a patient a 30-minute warning and percentage likelihood of a seizure.
An alarm would be triggered between 30 and five minutes before a seizure onset, giving patients time to find a safe place, reduce stress or initiate an intervention strategy to prevent or control the seizure.
Dr Kavehei said an advantage of their system is that is unlikely to require regulatory approval, and could easily work with existing implanted systems or medical treatments.
The algorithm that Dr Kavehei and team have developed can generate optimised features for each patient. They do this using what is known as a ‘convolutional neural network’, that is highly attuned to noticing changes in brain activity based on EEG readings.
Other technologies being developed typically require surgical implants or rely on high levels of feature engineering for each patient. Such engineering requires an expert to develop optimised features for each prediction task.
An advantage of Dr Kavehei’s methodology is that the system learns as brain patterns change, requiring minimum feature engineering. This allows for faster and more frequent updates of the information, giving patients maximum benefit from the seizure prediction algorithm.
The next step for the team is to apply the neural networks across much larger data sets of seizure information, improving sensitivity. They are also planning to develop a physical prototype to test the system clinically with partners at the University of Sydney’s Westmead medical campus.
Source: Marcus Strom – University of Sydney
Publisher: Organized by NeuroscienceNews.com.
Image Source: NeuroscienceNews.com image is in the public domain.
Original Research: Abstract for “Convolutional neural networks for seizure prediction using intracranial and scalp electroencephalogram” by Nhan Duy Truong, Anh Duy Nguyen, Levin Kuhlmann, Mohammad Reza Bonyadi, Jiawei Yang, and Omid Kavehei in Neural Networks. Published May 7 2018.
Convolutional neural networks for seizure prediction using intracranial and scalp electroencephalogram
Seizure prediction has attracted growing attention as one of the most challenging predictive data analysis efforts to improve the life of patients with drug-resistant epilepsy and tonic seizures. Many outstanding studies have reported great results in providing sensible indirect (warning systems) or direct (interactive neural stimulation) control over refractory seizures, some of which achieved high performance. However, to achieve high sensitivity and a low false prediction rate, many of these studies relied on handcraft feature extraction and/or tailored feature extraction, which is performed for each patient independently. This approach, however, is not generalizable, and requires significant modifications for each new patient within a new dataset. In this article, we apply convolutional neural networks to different intracranial and scalp electroencephalogram (EEG) datasets and propose a generalized retrospective and patient-specific seizure prediction method. We use the short-time Fourier transform on 30-s EEG windows to extract information in both the frequency domain and the time domain. The algorithm automatically generates optimized features for each patient to best classify preictal and interictal segments. The method can be applied to any other patient from any dataset without the need for manual feature extraction. The proposed approach achieves sensitivity of 81.4%, 81.2%, and 75% and a false prediction rate of 0.06/h, 0.16/h, and 0.21/h on the Freiburg Hospital intracranial EEG dataset, the Boston Children’s Hospital-MIT scalp EEG dataset, and the American Epilepsy Society Seizure Prediction Challenge dataset, respectively. Our prediction method is also statistically better than an unspecific random predictor for most of the patients in all three datasets.