Improving the detection of mTBI via Complexity Analysis by adopting an appropriate symbolization technique. A Magnetoencephalography Resting – State Study

Diagnosis of mild Traumatic Brain Injury (mTBI) was difficult due to the variability of obvious brain lesions using magnetic resonance imaging (MRI) or computed tomography (CT) scans. A promising tool for exploring potential biomarkers for mTBI is magnetoencephalography which has the advantage of high spatial and temporal resolution. By adopting proper analytic tools from the field of symbolic dynamics like Lempel-Ziv complexity (LZC), we can objectively characterize neural network alterations compared to healthy control. LZC is an estimator of the complexity of the system by enumerating the different patterns of the sequence. LZC needs first to binarize the time series using mean amplitude as the threshold. This procedure oversimplifies the rich information of brain activity captured via MEG. For that reason, we adopted neural-gas (NG) algorithm which has already been used for multichannel common symbolization. NG can transform a time series into more than two symbols by learning brain dynamics with a small error. To compare LZC with the NG symbolization approach, we adopted a proper complexity estimator called complexity index (CI). The whole analysis was presented to magnetoencephalographic (MEG) recordings of 30 mild Traumatic Brain Injury (mTBI) patients and 50 normal controls in δ frequency band. We compared CI and LZC via a classification procedure with k-NN and Support Vector Machines. Our results demonstrated that mTBI patients could be separated from normal controls with more than 97% classification accuracy based on CI with highest values considering to right frontal areas. In addition, a reversal relation between complexity and transition rate was demonstrated for both groups. These findings indicate that symbolization complexity could have a significant predictive value in the development of reliable biomarkers to help with the early detection of mTBI. Keywords—component; complexity; MEG; mTBI; Symbolization; Lempel-Ziv; Symbolic Dynamics