Presenting a new search strategy to select synchronization values for classifying bipolar mood disorders from schizophrenic patients

There is a growing interest to employ synchronization methods to reveal natural connections among the brain lobes by measuring co-activations among EEG channels. Regarding high number of EEG channels, lots of synchronization indexes are determined between two by two channels leading to construct a high dimensional feature vector for each time frame. The objective of this paper is to propose an effective feature selection method to find discriminative synchronization indexes in order to classify patients with schizophrenia from those with bipolar mood disorder (BMD). The state-of-art synchronization methods from various domains such as phase-locking value (PLV), robust synchronization (RS), and synchronization likelihood (SL), were implemented to provide a rich feature set in order to classify the two groups. To increase the classification accuracy, a capable feature selection scheme is proposed entitled greedy overall relevancy (GOR) to select discriminative synchronization indexes. The elicited synchronization vectors of 53 subjects imposed to support vector machine (SVM) classifier and the classification result with and without applying GOR, provided 92.45% and 88.68% accuracy, respectively. Across-group variance (AGV) is chosen as a rival method to GOR; the selected features by AGV entered to the classifier resulting in 91.34% accuracy. Using pair T-test exhibits the significant superiority of GOR to AGV such that P-value determined less than 0.05. To the best of authors' knowledge, this is the first attempt to utilize the selected synchronization indexes as informative features applying to a classifier for diagnosing the psychiatric patients.