Single trial behavioral task classification using subthalamic nucleus local field potential signals

Abstract

Deep Brain Stimulation (DBS) has been a successful technique for alleviating Parkinson's disease (PD) symptoms especially for whom drug therapy is no longer efficient. Existing DBS therapy is open-loop, providing a time invariant stimulation pulse train that is not customized to the patient's current behavioral task. By customizing this pulse train to the patient's current task the side effects may be suppressed. This paper introduces a method for single trial recognition of the patient's current task using the local field potential (LFP) signals. This method utilizes wavelet coefficients as features and support vector machine (SVM) as the classifier for recognition of a selection of behaviors: speech, motor, and random. The proposed method is 82.4% accurate for the binary classification and 73.2% for classifying three tasks. These algorithms will be applied in a closed loop feedback control system to optimize DBS parameters to the patient's real time behavioral goals.

An FFT-based synchronization approach to recognize human behaviors using STN-LFP signal

“…It has been shown [2,3] that β frequency range (13-30Hz) of LFP signals is an appropriate feature to discriminate different human behavior in the time-frequency domain. Therefore, we used the complex Morlet wavelet, which proved to be a suitable method for biomedical signal processing, to calculate the spectrogram of the raw LFP signals [2,15,17]. For each trial the wavelet coefficients in the β frequency range, calculated inside the [-1, 1] seconds interval around the onset, was used as the feature vector.…”

“…The effect of the FFT-based synchronization on the human behavior classification was evaluated using two recently proposed SVM [15] and MKL [17] approaches. We also studied the effect of various kernel functions on the classification performance, including linear K(x, y) = x T y+c, S RS AM MM BP 53 12 11 12 12 S 9 60 10 2 19 RS 12 11 57 7 13 AM 10 1 3 80 6 MM 15 19 11 2 53 Table II.…”

“…Santaniello et al, [14] presented a closed-loop DBS system capable of adjusting the stimulation amplitude using the LFP feedback from Ventral Intermediate Nucleus (VIM) of the thalamus. The time-frequency analysis of LFPs has been considered to classify different human behavioral activities [15][16][17]. Several classification methods based on Hidden Markov Model (HMM) and Deep Neural Network have been suggested in [18][19][20].…”

Long-Term Task- and Dopamine-Dependent Dynamics of Subthalamic Local Field Potentials in Parkinson’s Disease

“…Although the number of simple behavioral tasks performed by subjects was limited, STN-LFP activity was sensitive to the particular type of motor program. Patterns within STN-LFP signals have been determined to be specific to different motor programs allowing for high classification rates [20]. The data presented demonstrate that STN-LFP activity may be an appropriate feedback signal that is stable over time, providing relevant patient-specific information on a subject’s behavior and medication level.…”

“…Local field potentials (LFP) recorded from the STN have the potential to be a robust control signal to indicate a change in a patient’s state [12,13,14,15]. STN-LFPs have been shown to correlate with a patient’s Parkinson’s disease symptom state [16], levodopa medication level [17,18], behavior [19,20,21] and neurostimulation intensity [8]. Furthermore, LFP signals are stable over long periods of time, as evidenced by recordings from the motor cortex [22] and STN [23,24], a necessary characteristic for a feedback signal in a closed-loop DBS system.…”