Electroencephalography-Based Assessment Of Severe Disorders of Consciousness

We investigated electroencephalography (EEG) signals for the assessment of patients with severe disorders of consciousness. The EEG signals elicited by different stimuli, such as calling by name and music, were collected from 20 patients, including 10 in a minimally conscious and 10 in a vegetative state. A classification-based assessment framework was employed, consisting of the following major components: a preprocessing step, applied to remove the noises in the EEG data; selection of two types of features, including common spatial patterns and partial least squares; and application of the multiple kernel support vector machine (SVM) algorithm to perform the training and classification. Our results indicate that the EEG features detected in severe disorders of consciousness are significant for different auditory stimuli, and the multiple kernel SVM yields the best classification performance among different classifiers. We achieved average classification accuracies of 81.24% and 94.60% for calling by name and musical stimuli, respectively. The proposed method uses EEG signals to effectively classify patients into minimally conscious and vegetative states. It could be a promising tool for quantitatively identifying states of severe disorder of consciousness, and could also provide auxiliary information for clinical assessment of the level of consciousness.