In a paper published on the preprint server Arxiv.org this week, IBM researchers describe SeizureNet, a machine learning framework that learns the features of seizures to classify various types. They say that it achieves state-of-the-art classification accuracy on a popular data set, and that it helps to improve the classification accuracy of smaller networks for applications with low memory and faster inference.

If the claims stand up to academic scrutiny, the framework could, for instance, help the over 3.4 million people with epilepsy better understand the factors that trigger their seizures. The World Health Organization estimates that up to 70% of people living with epilepsy could live seizure-free if properly diagnosed and treated.

SeizureNet is a machine learning framework consisting of individual classifiers (specifically convolutional neural networks) that learn the features of electroencephalograms (EEGs) — i.e., tests that evaluate the electrical activity in the brain — to predict seizure types. (It’s the evolution of a seizure classification system that IBM first revealed last February.) Sub-networks used data sampled from different frequency and temporal resolutions, with the best performance achieved when the test data was processed with a sampling rate of 96Hz.

The researchers trained SeizureNet using the world’s largest corpus of seizure recordings — Temple University’s TUH EEG Seizure Corpus — which contains 2,012 seizures. They excluded myoclonic seizures because of the small number of seizures recorded. For validation purposes, they divided seizures for each type proportionally into training and test sets.

VB Transform 2020 Online - July 15-17. Join leading AI executives: Register for the free livestream.

They report that SeizureNet achieved 98.4% classification accuracy — about 4 percentage points higher than the next best seizure detection model. Moreover, in an experiment, they showed that SeizureNet could be used to improve the generalization performance of smaller networks. One small model required 45 times fewer training parameters, 1,100 times less number of FLOPS, and 45 times faster inference after training.

“Automatic classification of epileptic seizure types in EEGs data can enable more precise diagnosis and efficient management of the disease,” wrote the coauthors. “This task is challenging due to factors such as low signal-to-noise ratios, signal artifacts, high variance in seizure semiology among epileptic patients, and limited availability of clinical data.”

IBM’s work comes after the publication of a paper by Australia’s Edith Cowan University and Pabna University of Science and Technology in Bangladesh in which researchers detailed a model that automatically classifies seizures using a two-step method. Other recent, similar efforts include a competition on Google’s Kaggle platform, sponsored by MathWorks, the National Institutes of Health’s National Institute of Neurological Disorders and Stroke (NINDS), and the American Epilepsy Society, that challenged participants to train algorithms on seizure sufferers’ EEG data.