Jun 30 2020
Scientists from Washington University in St. Louis’ McKelvey School of Engineering have devised a more efficient method to identify and precisely detect an epileptic seizure in real time by combining systems theory with artificial intelligence (AI).
This gif was recorded during two seizures, one at 2950 seconds, the other at 9200 seconds. The top left animation is of EEG signals from three electrodes. The top right is a map of the inferred network. The third animation plots the second eigenvalue, the single value used to detect seizures using the network inference technique. Image Credit: Washington University in St. Louis.
The researchers’ results were published in the Scientific Reports journal on May 26th, 2020.
The study was performed in the laboratory of Jr-Shin Li, a professor in the Preston M. Green Department of Electrical and Systems Engineering. It was led by Walter Bomela, a postdoctoral fellow in Li’s laboratory.
Other researchers included Shuo Wang, a former student of Li and currently an assistant professor at the University of Texas at Arlington, and also Chun-An Chou from Northeastern University.
Our technique allows us to get raw data, process it and extract a feature that’s more informative for the machine learning model to use. The major advantage of our approach is to fuse signals from 23 electrodes to one parameter that can be efficiently processed with much less computing resources.
Walter Bomela, Postdoctoral Fellow, Department of Electrical and Systems Engineering, Washington University in St. Louis
In the field of brain science, the present interpretation of a majority of the seizures is that they happen when normal activity of the brain is disrupted by a powerful and spontaneous hyper-synchronized firing of a group of neurons.
At the time of a seizure, if an individual is attached to an electroencephalograph—a device called an EEG that quantifies electrical output—the abnormal activity of the brain is denoted as augmented spike-and-wave discharges.
“But the seizure detection accuracy is not that good when temporal EEG signals are used,” added Bomela. The researchers eventually created a network inference method to enable the detection of a seizure and locate its site with enhanced precision.
During an EEG session, an individual has electrodes fixed to different areas on his or her head; each electrode records the electrical activity around that specific area.
We treated EEG electrodes as nodes of a network. Using the recordings (time-series data) from each node, we developed a data-driven approach to infer time-varying connections in the network or relationships between nodes.
Walter Bomela, Postdoctoral Fellow, Department of Electrical and Systems Engineering, Washington University in St. Louis
Rather than looking only at the EEG data—that is, the strengths and peaks of individual signals—the network inference method takes relationships into consideration. “We want to infer how a brain region is interacting with others,” Bomela added. The network is formed by the sum of these relationships.
As soon as a network is available, its parameters can be quantified holistically. For example, rather than quantifying the strength of one signal, the total network can be assessed for strength. The Fiedler eigenvalue is one parameter that is specifically useful. “When a seizure happens, you will see this parameter start to increase,” explained Bomela.
With regard to network theory, the Fiedler eigenvalue is further associated with a synchronicity of the network—if the value is bigger, the network is more synchronous. According to Bomela, “This agrees with the theory that during seizure, the brain activity is synchronized.”
Background noise and artifact can also be eliminated if there is a bias toward synchronization. For example, if individuals scratch their arms, the corresponding activity of the brain will be recorded on certain EEG channels or electrodes but it will not be synchronized with seizure activity.
In this manner, this kind of network structure intrinsically decreases the significance of irrelevant signals; only the synchronized brain activities will increase the Fiedler eigenvalue considerably.
This method presently works for a single patient. The following step is to incorporate machine learning to simplify the method for detecting varying forms of seizures across various patients.
The aim is to leverage the numerous parameters that define the network and apply them as traits to train the machine learning algorithm.
Bomela compares the way this will work to facial recognition software, which quantifies different traits—such as lips, eyes, etc.—generalizing from such examples to identify any kind of face.
The network is like a face. You can extract different parameters from an individual’s network—such as the clustering coefficient or closeness centrality—to help machine learning differentiate between different seizures.
Walter Bomela, Postdoctoral Fellow, Department of Electrical and Systems Engineering, Washington University in St. Louis
The reason for this is that in network theory, similarities in particular parameters are related to certain networks. Such networks, in this case, will correlate to different forms of seizures.
In future, an individual suffering from a seizure disorder can wear a device that is similar to an insulin pump. When the neurons start to synchronize, the device will send electrical interference or drugs to halt the seizure in its tracks. But before this can occur, a deeper understanding of the neural network is required.
“While the ultimate goal is to refine the technique for clinical use, right now we are focused on developing methods to identify seizures as drastic changes in brain activity. These changes are captured by treating the brain as a network in our current method,” Li concluded.
Journal Reference:
Bomela, W., et al. (2020) Real-time Inference and Detection of Disruptive EEG Networks for Epileptic Seizures. Scientific Reports. doi.org/10.1038/s41598-020-65401-6.
Source: https://wustl.edu/