Towards High Accuracy Classification of MER Signals for Target Localization in Parkinson’s Disease Ruben-Dario Pinzon-Morales, Member, IEEE,Alvaro-Angel Orozco-Gutierrez, Hans Carmona-Villada, Cesar-German Castellanos Abstract— In recent years Microelectrode recording (MER) analysis has proved to be a powerful localization tool of basal ganglia for Parkinson disease’s treatment, especially the Subthalamic Nucleus (STN). In this paper, a signal-dependent method is presented for identification of the STN and other brain zones in Parkinsonian patients. The proposed method, refereed as optimal wavelet feature extraction method (OWFE), is constructed by lifting schemes (LS), which are a flexible and fast implementation of the wavelet transform (WT). The operators in the LS are optimized by means of Genetic Algorithms and Lagrange multipliers considering information contained in MER signals. Then a basic Bayesian classifier (LDC) is used to identify STN and other types of basal ganglia nuclei. The proposed method introduced several advantages from similar works reported in literature. First, the method is signal-dependent and non a priori information is required to decompose the MER signal. Second, the classification accuracy is mostly depended on the feature selection stage because it is not enhanced by elaborated classifiers such as support vector machines or hidden Markov models. Finally, the generalization property of the OWFE has been validated with two databases and different types of classifiers such as k-NN classifier and quadratic Bayesian classifier (QDC). Results have shown that proposed method is able to identify the STN with average accuracy superior than 97%. I. I NTRODUCTION Microelectrode recording (MER) signal analysis has proved to be a powerful localization tool of basal ganglia for Parkinson disease’s treatment in recent years, especially the Subthalamic Nucleus (STN). MER recordings are strong non–stationary signals that represent the nonlinear electrical activity generated by neurons. They are mainly composed of three electrical phenomena: spike activity, local field potentials (LPF), and background electrical noise. These phenomena have been typically studied in three different approaches. The first and most common approach is temporal domain analysis, where spike-related variables are computed from MER signals. For instance, Chan et al. [1] extracted spikes by means of the wavelet transform (WT) and then computed the multidimensional Teager Operator to discrim- inate neural information from the STN and the Substantia Nigra pars Reticulata (SNr) nucleus. However, such methods could decline in performance when low spike activity nuclei are analyzed. For example, serious problems could arise in the Zone Incerta (ZI). The second approach is frequency R. D. Pinzon, M. and A. A. Orozco, G. are with the Tecnological University of Pereira, Colombia. {rdpinzonm, aaog}@utp.edu.co H. Camona, V. MD. Neurosurgeon, Neurocentro, Pereira, Colombia. C. G. Castellanos is with the Universidad Nacional de Colombia Sede Manizales. domain analysis where the low frequency components of the MER signal, refereed as LFP, are analyzed in order to extract patterns. For example, Michmizos et al. [2] managed to model the STN nucleus by computing the power spectral density (PSD) of the LFP signal in abnormal health states such as Parkinson’s disease. Although this approach has been proved to be a powerful indicator of zones inside the STN, outside findings still inconclusive. Finally, the last approach is a combination called time-frequency analysis. In these, MER signals are transformed into a different space, i.e, wavelet space [3], Short-time Fourier Transform (STFT) space [4], Gaussian mixture models (GMM) space, and others [5]–[7]. As result, hidden information can be revealed and used for nuclei discrimination. For example, Zaidel et al. [6] presented a dynamic identification of the boundaries of the STN nucleus based on the root mean square (RMS) and the PSD of the MER signal. In previous work, Pinzon et al. [7] has presented an identification method of various basal ganglia nuclei by means of the Hilbert-Huang Transform (HHT) and Hidden Markov Models (HMM) using MER signals. In spite of, results have shown acceptable classifica- tion rates, the empirical nature of the HHT and the elevated computational load of the HMM represent issues to solve. Here a high-accuracy method for identification of basal ganglia nuclei is presented. The core of the proposed method relies in the feature selection stage OWFE. The OWFE is an ensemble of signal-adapted wavelet decomposition, constructed by Lifting Schemes (LS), which extract highly discriminant information from MER signals. Then basic machine learning methods, including the linear Bayesian classifier (LDC), can be used as classifier. This work is organized as follows. In Sec. II MER signal databases are described and the OWFE is introduced. Then in Sec. III experimental results and discussions are given. Finally, in Sec. IV concluding remarks are drawn and future work presented. II. METHODS AND MATERIALS A. Database Intra–operative acquisitions were made on unmedicated awake patients that underwent deep brain stimulator im- plantation. Four patients aging 55 ± 6 (4 male/ 1 female) who signed informed consent participated. Microelectrode recordings were made using the ISIS MER System (Inomed Medical GmbH). MER signals were labeled by specialists in neurosurgery and neurophysiology. These procedures were performed by the Institute of Parkinson and Epilepsy of the