Abstract
In the sub-acute phase, mild Traumatic Brain Injury (mTBI) patients often experience post-concussion syndrome (PCS), which is associated with changes in neural networks. The purpose of this paper is the introducing of an optimal predictive machine learning approach for analyzing brain functional connectivity changes after mTBI using EEG data to diagnose and predict mTBI patients who experience PCS during the sub-acute phase. After pre-processing and extracting four brain signal frequency bands, functional connectivity metrics such as Phase Locking Value (PLV) and Phase Lag Index (PLI) are extracted for each frequency band. Based on the extracted features, a graph-based machine-learning model is applied to classify mTBI with PCS, mTBI without PCS and control samples, of which 74 subjects (32 controls and 42 mTBI) participated in this study, and 20 subjects from mTBI had PCS symptoms after six months. Particularly, metaheuristic algorithms are used to improve the classification performance, by exploring and selecting effective graph-based features. The results have shown that the proposed approach for analyzing graph-based features from the functional connectivity matrices is a suitable criterion for diagnosing and predicting PCS after mTBI. In addition, by selecting the effective features, the accuracy of the classification process improved significantly to about 97%.