Classifying deep brain neuronal activities by bursting parameters

A method for classifying neuronal activities from the deep brain nuclei, sub-thalamic nucleus (STN) and subtantia nigra (SNr) is proposed in this paper. Seven bursting relevant parameters, firing rate (FR), burst index (BI), pause index (PI), burst number (BN), spike number (SN), burst strength (BS) and spike strength (SS) were applied to analyze 54 trials of data from Parkinson's patients. Based on the statistical analysis, PI, BN, SN and SS showed significant difference between STN and SNr signals. Thus, the 4 significant parameters were further applied to construct a classifier by principal components analysis (PGA) and support vector machine (SVM). The effect of constructing a SVM classifier with and without performing PCA was also tested. Applying PCA which can transform parameters to orthogonal variables improved the accuracy rate of classification for 22% on average. The number of principal components used to develop the classifier was also assessed. Including the first 2 principal components obtained the best accuracy rate of classification in this study.