Presenting a Spatial-Geometric EEG Feature to Classify BMD and Schizophrenic Patients

Abstract

Schizophrenia (SZ) and bipolar mood disorder (BMD) patients demonstrate some similar signs and symptoms; therefore, distinguishing those using qualitative criteria is not an easy task especially when these patients experience manic or hallucination phases. This study is aimed at classifying these patients by spatial analysis of their electroencephalogram (EEG) signals. In this way, 22-channels EEG signals were recorded from 52 patients (26 patients with SZ and 26 patients with BMD). No stimulus has been used during the signal recording in order to investigate whether background EEGs of these patients in the idle state contain discriminative information or not. The EEG signals of all channels were segmented into stationary intervals called “frame” and the covariance matrix of each frame is separately represented in manifold space. Exploiting Riemannian metrics in the manifold space, the classification of sample covariance matrices is carried out by a simple nearest neighbor classifier. To evaluate our method, leave one patient out cross validation approach has been used. The achieved results imply that the difference in the spatial information between the patients along with control subjects is meaningful. Nevertheless, to enhance the diagnosis rate, a new algorithm is introduced in the manifold space to select those frames which are less deviated around the mean as the most probable noise free frames. The classification accuracy is highly improved up to 98.95% compared to the conventional methods. The achieved result is promising and the computational complexity is also suitable for real time processing.