Epilepsy affects millions of people in the U.S. (approximately three million in 2015, according to Healthline). It’s commonly diagnosed by interpretation of electroencephalograms, or EEGs — measurements of the brain’s electrical activity taken from the scalp. But the signals tend to be quite long. This makes them challenging to interpret.
Researchers at Edith Cowan University in Australia and Pabna University of Science and Technology in Bangladesh propose a solution in a newly published preprint paper on Arxiv.org (“Epileptic seizure classification using statistical sampling and a novel feature selection algorithm“): an artificially intelligent (AI) system that automatically classifies seizures using a two-step method.
It’s a relatively novel idea. Recent similar efforts include a competition sponsored by MathWorks, the National Institutes of Health’s National Institute of Neurological Disorders and Stroke (NINDS), and the American Epilepsy Society on Google’s Kaggle platform that challenged participants to train algorithms on seizure suffers’ EEG data. Scientists at IBM, meanwhile, earlier this year described a fast, highly accurate AI-assisted seizure classification system.
But the researchers in this study sought to design a data preprocessing technique that didn’t adversely impact accuracy. They claim that their approach, which involves reducing the size of both EEG data and the number of features (i.e., characteristics), outperforms some of the state-of-the-art methods for seizure detection.
“The objective is to identify the effectiveness of the data reduction using the representative sample data based on the sampling technique,” the paper’s authors wrote.
The team sourced benchmark EEG data from the University of Bonn — five datasets containing a hundred individual channels 23.6 seconds in length for a total of 4097 data points per channel — and isolated signals from ten participants total (five healthy patients and five who had epilepsy). Next, they calculated the optimal sample size, divided the EEG into different 5.9-second segments, estimated the sample size again for each segment, and merged the segments before applying an algorithm to extract 15 different features for a total of 60 per signal.
The researchers then divvied up the datasets into three categories — “healthy,” “interictal” (the period between seizures), and “seizure” — and used three different cases for classification (“healthy versus seizure,” “interictal versus seizure,” and “healthy and interictal versus seizure”). They evaluated five different machine learning algorithms for epilepsy classification — random forest, naive Bayes, support vector machine, k-nearest neighbor, and logistic model trees — and in tests found that the random forest classifier achieved the highest accuracy. Even with a reduction in data points as high as 30 percent, it had a confidence level of 95 percent.
“The experimental results show that our proposed method using sampling technique and feature selection algorithm along with the random forest classifier can be an effective solution for epileptic seizure classification,” they wrote.