WIXLIB

Clinical StudyStereotact Funct Neurosurg 2006;84:28–34DOI: 10.1159/000093720Published online June 1, 2006Automatic Microelectrode Recording Analysis andVisualization of the Globus Pallidus Interna andStereotactic TrajectoryJon Haakon Falkenberg a, b James McNames a Kim J. Burchiel baBiomedical Signal Processing Laboratory, Electrical and Computer Engineering Department,Portland State University, and b Department of Neurological Surgery, Oregon Health and Science University,Portland, Oreg., USAKey WordsDeep brain stimulation Globus pallidus interna Microelectrode recording analysis Parkinson’s disease Stereotactic surgeryAbstractLocating deep brain neuronal structures is required to accurately place deep brain stimulation (DBS) electrodes duringstereotactic surgery in patients with Parkinson’s disease andother movement disorders. This study investigates the efficacy of automatic microelectrode visualization and analysismethods to help neurosurgeons locate target structuresmore objectively, consistently, and easily during surgery. Tenpatients (4 males and 6 females) who underwent bilateralimplantation of DBS electrodes in the globus pallidus interna (Gpi), from 2001 to 2003, at the Oregon Health and Science University and the Portland Veterans AdministrationMedical Center were included. We compared the efficacy ofthe microelectrode recording signal energy, power spectraldensity (PSD), marginal probability density (mPDF), autocorrelation function (ACF), and partial ACF. mPDF and PSD estimates most accurately indicated the borders of the GPi target structure.Copyright 2006 S. Karger AG, Basel 2006 S. Karger AG, Basel1011–6125/06/0841–0028 23.50/0Fax 41 61 306 12 34E-Mail karger@karger.chwww.karger.comAccessible online at:www.karger.com/sfnIntroductionThe internal segment of the globus pallidus (GPi) andthe subthalamic nucleus (STN) are both targets of deepbrain stimulation (DBS) for the treatment of advancedParkinson’s disease (PD). Several clinical studies havecompared the two sites to determine which site providesgreater benefits to PD patients [1, 2]. Despite confirmedclinical benefits of Gpi- and STN-DBS, the ideal targetlocation in humans is not known. There are a number ofnon-standard assessment procedures, which makes comparisons between STN and GPi targets very difficult toevaluate [3]. One of the challenges in comparing STN andGPi targets is that methods for locating the intended DBStarget are dependent on subjective human judgment.There are currently no methods which define the optimal pallidal surgical or DBS site that yields minimalcomplications and maximal efficacy. Spike train analysisis a common method used to determine the GPi target [4,5]. However, spike train analysis introduces subjectivitybecause the investigator typically needs to make choicesin selecting which spikes to include in the analysis, and inavoiding artifact and non-stationary transients in the data[6]. In addition, during surgery the investigator uses additional cues (e.g. change in background noise) to determine the target and transitions from one neuronal structure to the next [7]. This is particularly true when multipleunits are firing simultaneously or when spikes are not dis-Kim J. Burchiel, MD, FACSDepartment of Neurological Surgery, Mail code L-472Oregon Health & Science University, 3181 SW Sam Jackson Park RoadPortland, OR 97239 (USA)Tel. 1 503 494 6207, Fax 1 503 494 7161, E-Mail burchiek@ohsu.edu

cernable. Consequently, there exists a clinical need tomake microelectrode recording (MER) analysis less subjective, more consistent and easier to locate targets.Patients and MethodsThis study included 10 consecutive PD patients (4 males and6 females) who underwent bilateral implantation of DBS electrodes in the GPi, from 2001 to 2003, at the Oregon Health andScience University and the Portland Veterans AdministrationMedical Center (PVAMC). Table 1 summarizes the patient information. One patient was not included due to technical problemswith data collection. All microelectrode trajectories were analyzed postoperatively.Surgical TechniqueEstablished surgical techniques were used, as described by Favre et al. [8].Microrecording TechniqueA microrecording system (Neurotrek, Alpha Omega Engineering, Nazareth, Israel) was used to acquire neurophysiologic signals at a sampling rate of 12 kHz. Tungsten bipolar microelectrodes with impedances from 0.05 to 0.72 M (measured at1,000 Hz) were used (table 2). Online visual and auditory analyseswere the primary methods for distinguishing the different structures along the trajectory and to select the target. Most MERswere at least 30 s in duration.Data AnalysisTen consecutive seconds of the recorded signal that deviatedthe least from the mean were selected for analysis per electrodedepth. Segments of 5 s or less were omitted and segments between5 and 10 s included.Energy/Rank EnergyThe energy of the segments was calculated by taking the standard deviations of the signal. The interpercentile range of the energy was calculated at 1–99, 5–95, 10–90 and 25–75% and medianenergy. Each energy measure was displayed at each of the different recording depths.Table 1. PD study patients: summary n, yearsInclusioncriteriaGPi 100GPi 101GPi 102GPi 103GPi 104GPi 106GPi 107GPi 108GPi 109GPi 110GPi 151726IP, DIDIP, TR, OOIP, DID, BRIP, DID, DYSIP, DID, OOIP, DID, OOIP, TR, DIDIP, TRIP, DIDIP, TR, DID, BRIP, OO, BR6618AverageIP Idiopathic PD; DID drug-induced kinesia; BR bradykinesia; OO on/off fluctuations; TR tremor; DYS dystonia.Table 2. Final target, impedance and acquired coordinates, arc and ring angleIDFinal targetImpedance, M Arc angle, Ring angle, leftrightleftrightleftrightleftrightGPi 100GPi 101GPi 102GPi 103GPi 104GPi 106GPi 107GPi 108GPi 109GPi 110GPi .679.680.682.376.0NA Not applicable; arc angle lateral/medial angle (read from the Leksell frame); ring angle anteroposterior angle with respect to the anterior commissure-posterior commissure line (read from the Leksell frame);blank cell not recorded.Automatic Microelectrode Recording ofthe Globus Pallidus InternaStereotact Funct Neurosurg 2006;84:28–3429

Fig. 1. MERs in a PD patient (GPi 103L) atdifferent depths along the stereotactic trajectory. Annotations ofGPe and GPi weremade during surgery. The same data werea subset of the data used to generate figure2.WM White matter, BU border unit.Power Spectral DensityThe frequency content of the selected data segment at eachdepth was estimated using power spectral density (PSD). A nonparametric estimation (Welch’s method) of PSD was applied tothe segmented signal [9]. Window length was 85.2 ms, and thesignal was divided equally into segments with 50% overlap. Eachsegment was windowed with a Hamming window. The modifiedperiodograms were computed and averaged. Different magnitudes of power were plotted as different colors. The color rangewas normalized between 0 and 99% of magnitude values.Partial Autocorrelation FunctionThe partial ACF was estimated at lags of 0.0833 ms over arange of 0–5 ms [10, 11]. Similarly to the aforementioned ACF, thedifferent coefficients were bounded by limits –1 and 1. The colorspectrum was set as a function of a hyperbolic tangent function.Marginal Probability Density FunctionThe distribution of the signal amplitudes of the data segmentat each depth was estimated using the marginal probability density function (mPDF). After the signal mean was removed, thevoltage amplitudes of the MER data were distributed over 100 binsand normalized with respect to the number of elements in thedata. Similarly to the PSD plot, the mPDF color range was set to0–99% of maximum magnitude.Extracellular MER data from 24 trajectories were recorded from 10 patients that underwent bilateral GPi implantation. Two patients had a DBS electrode placed without undergoing second-side MER (table 1). The averagerecording time for each of the trajectories was 36 8 15min (STD; n 16). Of the 10 patients, 3 patients (GPi 101,GPi 107, and GPi 108) received more than one pass on thesame hemisphere. The second pass was placed 2 mm medially to the initial trajectory, and the third pass in GPi101 was placed 4 mm laterally to the initial trajectory. Using the coordinates from the Leksell stereotactic frame,the lateral/medial angle was described as the arc angleand the anteroposterior angle with respect to the anteriorcommissure-posterior commissure line has been described as the ring angle (table 2).Representative cell discharges along the GPi trajectoryare shown in figure 1. The traces represent a subset of theraw microelectrode data collected from the left hemisphere (GPi 103L).Results of the different statistical signal processingand visualization methods are demonstrated in figure 2.Time SeriesThe power at each time point of the signal segment was estimated (time series). The instantaneous power was calculated bysquaring the signal. The smoothed instantaneous power signals(cutoff at 1 Hz) were decimated to 200 samples and displayed in10-second segments per electrode depth. The time series magnitude was normalized to a color range that was set to 0–99% of themaximum magnitude.Autocorrelation FunctionThe biased autocorrelation function (ACF) had a user-specified lag of 12 ms that produced a vector of 144 coefficients for eachrecording depth [10, 11]. Different amplitudes of the coefficientswere plotted as different colors. The color spectrum was setbounded by limits –1 and 1. A hyperbolic tangent function wasapplied to emphasize deviations from 0.30Stereotact Funct Neurosurg 2006;84:28–34ResultsFalkenberg /McNames /Burchiel

Fig. 2. An example of statistical properties of MER visualizationversus electrode depth of a patient with PD. Visualization of theentire MER trajectory (48 depths). The top plot (energy) represents annotations that were made during surgery. The next plot(PSD) shows the energy of the spontaneous neuronal discharge atdifferent depths along the stereotactic trajectory, and the rankenergy of the energy; from top to bottom 1–99th, 5–95th,10–90th, 25–75th percentiles, respectively. The PSD, mPDF andtime series (TS) domain plots show amplitudes ranging from low(blue) to high (red). The target structure (GPi) had a nominal target depth of 27.5 mm. The visualization clearly shows the boundaries of the target structure between 20.5 and 26 mm.PACF Partial ACF.Forty-eight different depths of MER data were collectedalong the stereotactic trajectory in this figure, of whichsome depths and structures are shown as raw data in figure 1.Different statistical and visualization methods werecompared to determine if there was one single visualization technique that would stand out in distinguishing theventral and dorsal borders of the GPi. The mPDF appearsto be able to distinguish the borders of the GPi the best(fig. 3).The neurosurgeon assigned confidence scores to howdistinguishable the dorsal and ventral border featureswere of the globus pallidus externa (Gpe) and GPi basedonline auditory and visual analyses (table 3). There wasno obvious correlation between the confidence levels andimpedance of electrodes (table 2), patient disease duration, patient age (table 1) and how readily visualizationand analysis techniques allowed for target detection(fig. 3).Automatic Microelectrode Recording ofthe Globus Pallidus InternaStereotact Funct Neurosurg 2006;84:28–34Globus Pallidus Externa and Border CellsThere were no obvious differences between the GPeand the border cells regarding the energy, PSD, mPDF,ACF or the partial ACF.31