Differences in functional network between focal onset nonconvulsive status epilepticus and toxic metabolic encephalopathy: application to machine learning models for differential diagnosis

Abstract

We aimed to compare network properties between focal-onset nonconvulsive status epilepticus (NCSE) and toxic/metabolic encephalopathy (TME) during periods of periodic discharge using graph theoretical analysis, and to evaluate the applicability of graph measures as markers for the differential diagnosis between focal-onset NCSE and TME, using machine learning algorithms. Electroencephalography (EEG) data from 50 focal-onset NCSE and 44 TMEs were analyzed. Epochs with nonictal periodic discharges were selected, and the coherence in each frequency band was analyzed. Graph theoretical analysis was performed to compare brain network properties between the groups. Eight different traditional machine learning methods were implemented to evaluate the utility of graph theoretical measures as input features to discriminate between the two conditions. The average degree (in delta, alpha, beta, and gamma bands), strength (in delta band), global efficiency (in delta and alpha bands), local efficiency (in delta band), clustering coefficient (in delta band), and transitivity (in delta band) were higher in TME than in NCSE. TME showed lower modularity (in delta band) and assortativity (in alpha, beta, and gamma bands) than NCSE. Machine learning algorithms based on EEG global graph measures classified NCSE and TME with high accuracy, and gradient boosting was the most accurate classification model with an area under the receiver operating characteristics curve of 0.904. Our findings on differences in network properties may provide novel insights that graph measures reflecting the network properties could be quantitative markers for the differential diagnosis between focal-onset NCSE and TME.

Inherent seizure susceptibility in patients with antihistamine-induced acute symptomatic seizure: a resting-state EEG analysis

“…The machine learning classifiers using the EEG measurements were applied to classify the output into one of the two groups. Our previous studies have provided descriptions of feature selection methods for the application of machine learning algorithms 18 – 20 . This study used the scikit-learn module for Python to implement various machine learning classification algorithms to differentiate the AIS and seizure-free groups, such as kernel Support Vector Machine (kernel SVM) 21 , k-Nearest Neighbor (k-NN) 22 , Random Forest (RF) 23 , Extreme Gradient Boosting (XGBoost) 24 , and Light Gradient Boosting Machine (Light BMG) 25 , 26 .…”