Recognition of Brain Structures from MER-Signals Using Dynamic MFCC Analysis and a HMC Classifier

Abstract

A novel methodology for the characterization of Microelectrode Recording signals (MER-signals) in Parkinson’s patients in order to recognize basal ganglia in the brain is presented in this work. The most common approach of MER signals analysis consists of time-frequency analysis through Short Time Fourier Transform, Wavelet Transform, or Filters Banks. We present an approach based on MEL-Frequency Cepstral Coefficients (MFCC) and K-means clustering to obtain dynamic features from MER-signals. A Hidden Markov Chain (HMC) with 1, 2, 3, and 4 states was used for the classification of four classes of basal ganglia: Thalamus (Tal), Zone Incerta (ZI), Subthalamic Nucleus (STN) and Substantia Nigra reticulata (SNr), achieving a positive identification over 82%. A performance analysis for each HHM model is presented using ROC curves.