Feature Extraction On Detecting Brain Functional States In Parkinson’s Disease

Deep brain stimulation (DBS) is an effective tool to alleviate and suppress motorrelated symptoms in Parkinson’s disease (PD). As a main surgery for advancedParkinsonian patients, the function of current DBS system is to deliver a train ofelectrical pulse with fixed frequencies and amplitudes to targeted nuclei. Throughmodifying the over-synchronization phenomenon in the basal ganglia network, DBScan significantly improve the condition of resting tremor, rigidity, bradykinesia. Inthat the mode of conventional DBS provides a continuous train of electrical stimulus,DBS offer a low efficacy and a series of side effects. Closed-loop DBS is beingdesigned to detect and classify the PD related brain functional states so that thesystem can promote the clinical efficacy and efficency by delivering on-demandstimulus. Taking advatange of better clinical effect and low power consumption,closed-loop DBS is becoming the next generation DBS.In this paper, we try to reveal the relationship between brain signals andParkinsonian brain functional states. By analyzing the local field potential (LFP)singals directly recorded from the subthalamic nucleus in PD patients, we appliedadvanced methods to extract the features inside LFP which can be biomarkersrepresenting the disease states so that the closed-loop DBS control system can give aprecise order of trigger time to the stimulator.Firstly, EMG and LFP signals related to Parkinsonian rest tremor are recordedand selected. We adopted power spectrum analysis to determine the the neuralactivities’ energy distribution in the frequency domain. The results show the power density in Delta（0-4Hz）、Theta（4-8Hz）、Alpha（8-12Hz）�'�Beta（12-30Hz）contains the most power in LFP.Secondly, we used time-frequency analysis to study the dynamic characteristicsof components in band3-30Hz,and we found3-20Hz band has a stronger correlationwith disease states. The results of short-time Fourier analysis to the components in3-20Hz demonstrate that only in tremor state, there are several harmonic componentsin Delta, Theta, Alpha and low Beta bands which has a strong synchronizationphenomenon.Thirdly, to explain the synchronization phenomenon mentioned above, Hilbertbased corss-frequency correlation analysis was applied to reveal linear couplinginteractions between amplitude signals of these harmonics in tremor state.ThenGranger causality method shows low frequency components has stronger causality tohigh frequency components which, to some extent, explain the mechanism of thecorrelation relation.Finally, we focus the nonlinear interations between these harmonic components in3-20Hz. High order statistics method can detect quadratic phase coupling in the LFPsignals. The features in distributions and intensity of phase coupling in tremor andresting states can classify the functional states of subthalamic nucleus.To sum up, in this paper we realized to classify the Parkinsonian subthalamicnucleus functional states by researching the linear and nonlinear interactions betweendifferent components in LFP signal.