Prediction of Consciousness Recovery in Deep Coma after Acute Traumatic Brain Injury based on EEG Functional Connectivity Features and Support Vector Machine Classifier

In recent years advanced neurocomputing and machine learning techniques have been used successfully for the prediction of medical outcomes. In the presented study, a novel computer model is presented for the prediction of the recovery of consciousness in patients with severe traumatic brain injuries in acute coma. The model reconstructs the time courses of synchronized neuronal components in EEG data and uses their topographies as features in Support Vector Machine classifier.
The data were obtained from patients with severe TBI in acute state (1-7 days after trauma). A total of 21 patients with GCS score of 3-4 upon arriving at the emergency unit were assigned to training and test sets. The core set of predictors included topographies of cross-frequency phase interactions and correspondent Phase Locking Values. With each of these sets we designed a model predicting the recovery of consciousness based on Glasgow Coma Scale measurements as of 3 months after the trauma. We performed a binary classification: ‘positive’ for those with GCS score > 10, and ‘negative’ for the patients with GCS < 7 being measured 3 months after the trauma.
Accuracy, ROC/AOC were used as measurements of performance of the predictive model. It was discovered that the patterns of EEG functional connectivity in the theta and alpha frequency bins recorded during tactile simulation in acute coma are distinctive with regard to the level of consciousness which the patients exhibited further on. In the area of predicting the recovery of consciousness in TBI patients, Gaussian kernel SVM classifier delivered modest results.
This study shows that nonlinear analysis of EEG can be a useful method for discriminating between positive and negative outcome with regard to recovery of consciousness in patients early after trauma. It is suggested that this analysis may be a complementary tool to help clinicians deliver early bedside diagnosis.