Evaluating Performance of EEG Data-Driven Machine Learning for Traumatic Brain Injury Classification.

Meijun Ye,Michael Caiola,Nicolas Vivaldi,Krystyna Solarana

doi:10.1109/tbme.2021.3062502

Abstract

Objectives:Big data analytics can potentially benefit the assessment and management of complex neurological conditions by extracting information that is difficult to identify manually. In this study, we evaluated the performance of commonly used supervised machine learning algorithms in the classification of patients with traumatic brain injury (TBI) history from those with stroke history and/or normal EEG.Methods:Support vector machine (SVM) and K-nearest neighbors (KNN) models were generated with a diverse feature set from Temple EEG Corpus for both two-class classification of patients with TBI history from normal subjects and three-class classification of TBI, stroke and normal subjects.Results:For two-class classification, an accuracy of 0.94 was achieved in 10-fold cross validation (CV), and 0.76 in independent validation (IV). For three-class classification, 0.85 and 0.71 accuracy were reached in CV and IV respectively. Overall, linear discriminant analysis (LDA) feature selection and SVM models consistently performed well in both CV and IV and for both two-class and three-class classification. Compared to normal control, both TBI and stroke patients showed an overall reduction in coherence and relative PSD in delta frequency, and an increase in higher frequency (alpha, mu, beta and gamma) power. But stroke patients showed a greater degree of change and had additional global decrease in theta power.Conclusions:Our study suggests that EEG data-driven machine learning can be a useful tool for TBI classification.Significance:Our study provides preliminary evidence that EEG ML algorithm can potentially provide specificity to separate different neurological conditions.

Highlights

T RAUMATIC brain injury (TBI) presents a significant challenge to civilian and military medicine
1) Models Trained With Features Selected By Statistics: In statistical analysis, rank sum and false discovery rate analyses identified 98 features out of the 1333 in the raw set (∼7.35%) and 82 in the clean set (∼6.15%) that were significantly different between traumatic brain injury (TBI) and normal subjects, including sex
When comparing the performance of models trained with truly labeled data and randomly labeled data with 10-fold cross validation (CV), all models trained with truly labeled performed significantly better than randomly labeled data with the only exception of Support vector machine (SVM) fine Gaussian at 10-10 significance level (SL)

Summary

Introduction

T RAUMATIC brain injury (TBI) presents a significant challenge to civilian and military medicine. According to the Centers for Disease Control and Prevention (CDC), an estimated 2.5 million people sustain a TBI annually, contributing to a third of all injury-related deaths in the United States. Given the high societal and economic costs of untreated TBI, it is recognized as a significant military and public health concern. Neurological Glasgow Coma Scale (GCS) is a clinical index universally used to classify TBI as mild, moderate or severe. CT scan is used to detect structural brain lesions. Though useful in the clinical management of TBI, these methods do not provide enough sensitivity to detect mild TBI and monitor the progression of TBI at different severities. Efforts are ongoing to seek for alternative clinical assessment tools for TBI, including body-fluid analysis, advanced imaging modalities (i.e., diffuse tensor imaging [DTI], positron emission tomography [PET]) and neurophysiological signals (i.e., eye movement and electroencephalography [EEG])

Objectives

Methods

Results

Discussion

Conclusion