Citation
Using local field potentials to differentiate brain regions and optimize target localization for deep brain stimulation surgery

Material Information

Title:
Using local field potentials to differentiate brain regions and optimize target localization for deep brain stimulation surgery
Creator:
Kolb, Rachel Vera ( author )
Language:
English
Physical Description:
1 electronic file (29 pages) : ;

Subjects

Subjects / Keywords:
Brain stimulation ( lcsh )
Neurology ( lcsh )
Brain stimulation ( fast )
Neurology ( fast )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Review:
Deep Brain Stimulation (DBS) is a common surgical treatment option for individuals with movement disorders like Parkinson's disease (PD), and requires precise electrode placement to electrically stimulate the target brain structure. Traditionally for DBS, electrode placement is determined based on non-invasive imaging techniques and the stereotactic coordinates are computed using software. To aid in more accurate DBS electrode placement than simply the computer generated location, I examined whether local field potentials (LFPs) obtained from the macroelectrode contacts could distinguish when the electrode is within the target structure because local field potentials give a more global recording perspective than single unit recordings. Several studies have indicated that increased power spectral activity in the beta band (12-40 Hz) in the STN has been associated with akinetic motor output in PD. To compare these results using LFPs, the power spectrum was analyzed in Matlab using standard signal processing techniques. It became clear that within the beta band there are signatures that distinctly separate the target STN from other structures along the trajectory. The LFP power spectrum showed increased amplitude and an increased number of oscillations in the beta band for depths confirmed to be in the middle of the STN. Analysis of the average relative power for specific depths identified along the trajectory indicated that power continuously increases from striatum to thalamus to STN. These features of LFPs within the beta band seem to be distinct markers of the STN and could be incorporated into future technologies to more confidently identify the target electrode location and validate the use of LFPs to monitor during DBS.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: Adobe Reader.
Statement of Responsibility:
by Rachel Vera Kolb.

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
982698074 ( OCLC )
ocn982698074
Classification:
LD1193.L56 2016m K65 ( lcc )

Downloads

This item has the following downloads:


Full Text
USING LOCAL FIELD POTENTIALS TO DIFFERENTIATE BRAIN REGIONS AND OPTIMIZE
TARGET LOCALIZATION FOR DEEP BRAIN STIMULATION SURGERY
by
RACHEL VERA KOLB B.S., The College of New Jersey, 2010
A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Masters of Science Bioengineering Program
2016


This thesis for the Master of Science degree by Rachel Vera Kolb has been approved for the Bioengineering program
by
Emily Gibson, Chair Gidon F el sen John Thompson
April 28th, 2016
ii


Kolb, Rachel Vera, MS, Bioengineering
Using Local Field Potentials to Differentiate Brain Regions and Optimize Target Localization for Deep Brain Stimulation Surgery
Thesis directed by Assistant Professor Emily Gibson.
ABSTRACT
Deep Brain Stimulation (DBS) is a common surgical treatment option for individuals with movement disorders like Parkinsons disease (PD), and requires precise electrode placement to electrically stimulate the target brain structure. Traditionally for DBS, electrode placement is determined based on non-invasive imaging techniques and the stereotactic coordinates are computed using software. To aid in more accurate DBS electrode placement than simply the computer generated location, I examined whether local field potentials (LFPs) obtained from the macroelectrode contacts could distinguish when the electrode is within the target structure because local field potentials give a more global recording perspective than single unit recordings. Several studies have indicated that increased power spectral activity in the beta band (12-40 Hz) in the STN has been associated with akinetic motor output in PD. To compare these results using LFPs, the power spectrum was analyzed in Matlab using standard signal processing techniques. It became clear that within the beta band there are signatures that distinctly separate the target STN from other structures along the trajectory. The LFP power spectrum showed increased amplitude and an increased number of oscillations in the beta band for depths confirmed to be in the middle of the STN. Analysis of the average relative power for specific depths identified along the trajectory indicated that power continuously increases from striatum to thalamus to STN. These features of LFPs within the beta band seem to be distinct markers of the STN and could be incorporated into future technologies to more confidently identify the target electrode location and validate the use of LFPs to monitor during DBS.
The form and content of this abstract are approved. I recommend its publication.
Approved: Emily Gibson
III


TABLE OF CONTENTS
CHAPTER
I INTRO...........................................................................1
What is Deep Brain Stimulation?.................................................1
DBS and Parkinsons Disease.....................................................2
Using Local Field Potentials to study DBS for PD................................3
II METHODS........................................................................4
DBS Surgery.....................................................................4
LFP Electrophysiology...........................................................4
Subjects & Sample Size..........................................................5
Matlab Analysis.................................................................5
III RESULTS......................................................................10
Increase in Amplitude and Number of Signal Oscillations for the Power Spectral Density of the
STN Within the Beta Band.......................................................10
Distinct Increase for Average Relative Power Emerges Across Multiple Patients..16
Average Relative Power Continuously Increased from Striatum to Thalamus to STN.17
IV DISCUS SION..................................................................20
V CONCLUSION...................................................................23
BIBLIOGRAPHY.....................................................................24
IV


CHAPTERI
INTRO
What is Deep Brain Stimulation?
The development of deep brain stimulation (DBS) as a therapeutic tool has been evolving since 1870 when Fritsch & Hitzing demonstrated localized electrical stimulation can be used to investigate brain function.1 Although it wasnt until the development of stereotaxic devices closer to 1957 that deeper stimulation of the brain began to be explored.11 In the 1960s the concept of high-frequency stimulation to treat movement disorders became popular and surgical lesions were targeted by implanting multiple electrodes together using wires and stimulating them for weeks at a time.111 DBS continued to be used for epilepsy, cerebral palsy and Parkinsons Disease (PD) treatment, and throughout the 1970slv and 1980sv. It wasnt until the 1990s when DBS and chronic stimulation technology from pacemakers became more prevalently accepted as a treatment option for a variety of disorders?1 Today, DBS is considered a reversible neurosurgical treatment that can be used for advanced movement disorders such as: dystonia, essential tremor, and PD. It is also an emerging therapy that has the potential to provide relief for treatment-resistant diseases such as obsessive-compulsive disorder, Tourettes, and other neuropsychiatric conditions?111 Different sites of stimulation within deep brain structures have been linked to ameliorating specific symptoms: essential tremor has been linked to the ventral intermediate thalamus, PD and dystonia have been linked to both the Globus pallidus interna (GPi) and subthalamic nucleus (STN)X, and recently Tourette syndrome has been linked to the GPi?1 DBS surgery involves the stereotactic implantation of a quadripolar macroelectrode into a deep brain structure to disrupt the abnormal neuronal firing rates/patterns to treat motor symptoms. DBS can have varying effects throughout structures in the brain due to characteristics like stimulation amplitude, temporal spacing, and properties of individual cells, stimulus field geometry, target geometry, and even the underlying pathophysiology of the disease state?11 The Federal Drug Administration of the US approved the first DBS device, Medtronics Activa DBS Therapy System, for essential tremor in 1997 and PD in 2002.xm
1


DBS and Parkinsons Disease
The motor symptoms of PD can be managed using medication, which increases or substitutes dopamine within the brain, commonly levodopa is used because it is a natural chemical, which is converted into dopamine once it passes into the brain. Over time, levodopa-based treatments become less effective at managing movement disorders, requiring higher dosages and can ultimately lead to dyskinesia. Patients with PD that are no longer responsive to pharmacological and nonpharmacological treatments can consider DBS as the next option to manage their disease symptoms. For PD DBS, targets can be either the STN or GPi to optimize patient outcome, but the most common target is the STN, a node within the basal ganglia circuit. Within the basal ganglia circuit, the STN receives excitatory inputs from the cerebral cortex, other basal ganglia components, thalamus, pedunculopontine nucleus and substantia nigra pars compacta and inhibitory inputs from Globus pallidus external. The STN has glutamatergic, excitatory outputs to Globus pallidus, substantia nigra pars reticulata (SNr) and pedunculopontine area.1 Targeting the STN has also been associated with reduced medication and less frequent battery changes because the target area is smaller, resulting in a better long-term economic profile.11 The STN target is initially located using preoperative non-invasive imaging and then informed using intraoperative electrophysiological techniques. Current surgical practice depends upon single-unit electrophysiology to define the optimal implant target. Single-unit potentials are highly accurate and taken from a small sample of neurons (6-10). They are highly susceptible to mechanical and electrical perturbation during recordings, greatly reducing their reliability. It is possible to collect local field potentials (LFPs) during DBS along the trajectory to the target from both the microelectrode and the DBS electrode without requiring any additional electrodes or hardware. LFPs represent the population-averaged signal of electrical current flowing in nearby neurons and give a larger sample of neurons (~1000)m These LFP signals can be recorded at a lower frequency than single-unit recordings, which results in signals less affected by electrode interface or local geometry. Electrical signal recordings obtained from DBS can be separated into frequency bands and then associated with relevant clinical signatures: delta (0-4 Hz), theta (4-8 Hz), alpha (8-12 Hz), beta (12-40 Hz), and gamma (40-100 Hz). LFP recordings in DBS
2


with an STN target from LastName et al., (2009) indicated that there was elevated LFP power within the beta frequency band linked to motor symptoms of PD, like rigidity and bradykincsia. Frequency modulation in the beta band is regulated by circulating dopamine levels which is consistent with PD because it is caused by the death of dopaminergic neurons. The STN is where information processing information from across the cortex converges, so the population-based metric of LFP is a more logical choice than the single-unit recordings to look at broader state changes in PD.
Using Local Field Potentials to study DBS for PD
If it is possible to identify distinct signal characteristics for specific regions along the
electrode trajectory LFPs could be utilized for determining the borders of different brain structures. The LFP signatures within the beta band of the STN could also be utilized as a biomarker to develop a closed-loop stimulation DBS device. LFPs are highly robust signals that represent an entire neuronal population and have the potential to aid in surgical targeting of specific brain regions during DBS. In this study, we examined whether LFP signals could be used to objectively predict entry and exit of STN in PD patients undergoing DBS. Distinct characteristics emerged within the beta band of the recorded LFPs to differentiate the STN from other structures along the surgical trajectory, including an increase in instantaneous power, increased number of peaks, and increased relative average power. This study confirmed that LFP can be utilized as a tool to enhance DBS and improve target localization for the STN.
3


CHAPTER II
METHODS
All procedures, data collection and analyses were approved by the Colorado Multiple Institution Review Board (COMIRB). All LFP analyses were conducted on de-identified data from patients who provided their voluntary and informed consent.
DBS Surgery
Data were collected during routine DBS implantation surgery (described in detail Abosch et al., 2013 (Stereotactic Functional Neurosurgery 91(1): 1-11)). Briefly, pre-operative brain scans both magnetic resonance (MR) and computed tomography (CT) with a contrast-enhancing head frame are imported into stereotactic planning software to derive initial STN target coordinates associated with a CRW (Cosman-Robert-Wells) stereotactic system. The stereotactic frame is aligned on the patient based on the planned trajectory from the computer software. Three microelectrodes are used to traverse the planned trajectory starting at 25 mm above target (ventral border of STN) and are spaced 2mm apart from each other. The microelectrodes are advanced in submillimetric steps (100-500 pm). High impedance recordings from the microelectrodes are displayed on an oscilloscope and used to visually assess location in the brain and ultimately verily the DBS target. Once the STN is identified, patients engage in arm and leg motions to drive activity in the STN associated with somatotopic territory. During microelectrode recordings, patients are awake and initiating arm and leg movements Once the microelectrodes are determined to be in the STN, a single lead is chosen to implant Following macrostimulation to assess therapeutic benefit and side effect profile, the DBS electrode is secured in place.
LFP Electrophysiology
The microelectrode on each lead and the DBS macroelectrode are capable of recording LFPs (1.3 kHz sampling frequency). The LFP recordings are taken in incremental steps (100-500 pm) along the trajectory for at least 10 seconds for each signal recording. The LFP signals are digitally sampled and exported using specialized software to Matlab.
4


Subjects & Sample Size
The obtained LFPs are from patients that underwent DBS surgery between 2014-2016 at University of Colorado Hospital. All subjects have been diagnosed with PD and the target implantation for DBS surgerys was the STN. A total of 29 LFP recordings were analyzed, with 23 used for the analysis. All raw data was imported into Matlab and viewed to check for any recording issues or outliers. Some patient files were not used for analysis due to abnormalities in the signal recordings stemming from noise and interference.
Matlab Analysis
All of the LFP recordings were sampled at 1.3 kHz prior to being imported into Matlab. The patient recordings were rim through a preprocessing code written in Matlab to remove the mean from each electrode and band pass the signals using a 4th order zero-phase lag filter at 350 Hz. Any data above 21mm from the target was removed due to proximity to an impedance check that could corrupt the data. The mean of the data was removed by evaluating each CLFP electrode separately for each depth data file, using the nanmean function in Matlab to find the average value of an array ignoring any nan values which was subtracted from the overall signal for each electrode at each depth. A digital filter was used to make sure that the Nyquist frequency rule was met, that the minimum sampling rate of the signal is twice the highest frequency present to prevent any introduction of errors. If the Nyquist frequency was above 350 Hz it was automatically set to 350 Hz for the passband filter. The designfilt function was used to design a low pass infinite impulse response (HR) filter with a 4th order and pass band frequency of 350 Hz based on the sampling frequency of the recording -1500 Hz. The HR is a commonly used digital filter that uses feedback to generate an infinite number of values for a single impulse. An HR filter uses less memory and requires less calculations than other digital filters, a 4th order filter was used based on desired specifications from the passband, stopband attenuation and transition widths. The data was indexed after the filtering to remove any recording depths preceding the impedance check. The resulting files were saved into a Matlab structure and saved to the local drive. The preprocessed data was then sent through a second preprocessing code written in Matlab designed to eliminate any depth recordings that have values more than six times the standard deviation
5


above the mean. This was done by removing any nan values in the depths resulting from removal in the first preprocessing round. For each recording depth the median of the signal was found using the median function and the interquartile range of the signal found by the iqr function. A threshold was calculated by adding the resulting median value with six times the interquartile range. The absolute value of the signal, obtained using the abs function, was compared to that threshold to determine if there was a recording error then that recording depth was not used. The resulting signals after this check were again saved as a structure to the local drive. Once both preprocessing steps were completed, the resulting data was analyzed. Prior to generating any figures all of the nan values were removed and the signal was multiplied by a bit resolution of 38.14 and then divided by 1000. For each depth, the first few moments of the recording, number of points as half of the sampling frequency, were eliminated to ensure that there was no residual noise from the movement of the electrode down the trajectory. The time axis was calculated for each depth recording based on the number of points in the data and the sampling frequency. All recordings were originally recorded for at least 10 seconds in the operating room, but due to preprocessing slightly decreased. To calculate the power spectral density (PSD) of the signal, the pwelch function from Matlab was used with a moving average filter using a window of 650 points and an overlap of 50 points. This function was chosen to calculate Welchs power spectral density estimate using an averaging estimator and because their power input scales each estimate of the PSD by the equivalent noise bandwidth of the signal. The resulting frequency output for the PSD calculation was then indexed to only include frequencies from l-50Hz. That same index was used on the PSD output to only obtain PSD values within the l-50Hz range. That resulting signal was converted into decibels by multiplying it by 10 log of base 10. To ensure that the recorded signals were all distinct for each depth, a magnitude squared coherence test was performed using the mscohere function, it finds the magnitude-squared coherence estimate using Welchs
6


average modified periodogram method.
A
B
Raw Data lfpi
---LFP2
---LFP3
3mm
Sagittal 11.0
10 ms
LFP1
LFP2
LFP3
Q) Signal Coherence
LFP1 v LFP2 LFP2 v LFP3 LFP1 v LFP3
0.25 -
5 10 15 20 25 30 35 40 45 50
Frequency (Hz)
Figure 1: (A) Schematic of surgical procedure set-up and trajectory. (B) Sample of time spectrum recording of three raw LFP signals. (C) Sample of power spectrum for three LFP signals including their standard deviation. (D) Sample signal coherence between all three LFP signal recordings indicating that each signal is distinct.
The data was entered into the function along with a 100-point hamming window with an overlap of 80, for frequencies under 100 at the given sampling frequency of ~1500Hz. To find the mean for one single depth across all electrodes the namneaiT function was again used for all three electrode means to obtain a single mean value for that depth. To calculate the maximum value for one specific depth the max function was used for all three electrodes to find a single maximum value at that depth. To calculate the mean power spectral density for one patient across all depths the TianmeaiT function was used again to find the average ignoring any nan values for each electrode individually
7


and then used again on an array with each electrodes mean across the entire frequency to calculate the average power for frequencies l-50Hz across all depths to use as a reference line. To calculate the number of peaks within specific bands, the frequency band of interest was indexed into the calculated power signal for that specific depth. That resulting power signal for the frequency band of interest was used as an input into the findpeaks function, which finds the values of the local maxima. The length of the resulting array was calculated using the length function to count the number of peaks. To analyze the distributions and variances of the number of peaks for each structure three preliminary statistical tests were done using functions within Matlab. The two sample Kolmogorov-Smimov test to determine if both samples are from the same continuous distribution using the kstest2 function, the Ansari-Bradley test to determine if both samples are from the same distribution using the ansaribradley function, and a two sample F-test to determine if both samples have normal distributions with the same variance using the vartest2 function. To generate the heatplot showing power across all patient depths from l-50Hz, power was calculated the same way as above. The power data was then sorted by depth using the sort function in Matlab to order the depths in descending order. Then any data rows that began with a nan value were removed using -isnan logic, to only leave values that were not a nan. To display the image, imagesc was used to visualize it with scaled colors, with an image property setting of CDataMapping to scaled to allow us to define the minimum and maximum color limits for the image to standardize across all electrodes as -25 to 8 dB. To compare multiple patients together and look at the power spectrum trends for the striatum, thalamus, and STN within the beta band each structure depth was approximated and the average power calculated. The average power was calculated the same way as described above for the following depth ranges, striatum 21 to 16 mm, space between striatum and thalamus 16 to 11 mm, thalamus 11 to 6 mm, STN 6 to 1 mm, and SNr 1 to -4mm. The average power for each region was calculated separately for each electrode and then all patients for that electrode were plotted together using the patchline function, then the average for all patients for each electrode was calculated by taking the mean and plotting it using the same patchline function. To precisely identify the structures for each patient specific to each electrode an excel fde was created that included the specific depths
8


that each structure was identified to be in from each surgery. The values found were defined to be the middle of the striatum, the top of the thalamus, the middle of the STN, and, where applicable, the top of the SNr. For each structure the power was calculated by electrode for each individual the same way as outlined above. In order to compare across all patients the average relative power was calculated by dividing the power for each specific structure by the maximum power found for that individual within that electrode. The median for each structure was calculated across all patients using the nanmedian function to ignore nan values for missing data. The 25th and 75th percentiles were calculated using the pretile function for the data. An ANOVA was run on the data to compare the striatum, thalamus, and STN using the anoval function. The SNr was not included in the ANOVA because there were not enough confirmed points within that structure.
9


CHAPTER III
RESULTS
Increase in Amplitude and Number of Signal Oscillations for the Power Spectral Density of the STN Within the Beta Band
Current literature has shown significant increases in the power spectral density amplitude within the STN for the beta band. In order to evaluate these previously published findings with our datasets, the multitaper power spectral density was calculated. The LFP power spectral density for four of the brain structures confirmed along the trajectory: striatum, thalamus, and STN is depicted in Figure 2. Each graph in Figure 2 includes LFP electrode recordings and the amplitude mean and maximum for that specific depth recording. It is clear that for the beta frequency band, 12-40 Hz, the power increases from striatum to thalamus and thalamus to STN. For the thalamus recording the LFP electrode traces are very close to the overall mean dashed line trace, and they clearly shift above that same line for the STN recording. The mean of the signal for the STN is higher for STN than for thalamus, 5.25 vs 4.93 dB. In Figure 3, between the thalamus and STN there also is a clear increase in oscillatory activity, specifically between 30-35 Hz. The thalamus has 2 local maxima whereas the STN has 5 local maxima. This figure supports the literature findings that the number of oscillations increase in the beta band for the STN compared the structures found higher along the trajectory like the striatum and thalamus. The analysis for the number of peaks and the statistical significance of the various tests run on the data to compare the striatum and thalamus to the STN can be seen in Table 1. The two-sample Kolmogorov-Smirnov test was used to determine if both samples are from the same continuous distribution, the Ansari-Bradley test was used to determine if both samples are from the same distribution, and a two sample F-test was used to determine if both samples have normal distributions with the same variance. The standard deviation for the STN is much lower than the striatum or thalamus and the mean and median is higher, indicating that there is less variation and larger values. The STN also has the smallest coefficient of variation for the three structures. The p-values found comparing the striatum and thalamus to the STN indicate that there are statistically
10


significant differences in the distribution between the STN and the striatum and that the values are not
from the same distributions.
11


12


Magnitude (dB) Magnitude (dB) Magnitude (dB)
LFP PSD for Striatum
Maximum = 9.02
J__________l__________l__________l_________l__________
15 20 25 30 35 40
Frequency(Hz)
LFP PSD for Thalamus
Maximum = 11.13
25 30
Frequency (Hz)
LFP PSD for STN


Figure 2: LFP Multitaper power spectral density in decibels for 12-40 Hz within the striatum, thalamus, and STN. Striatum recording has a mean of 1.93 dB, maximum of 9.02 dB, thalamus recording has a mean of2.63 dB, maximum of 11.13 dB, STN has a mean of 4.80 dB, maximum of 10.16 dB. The mean increases consistently from striatum to thalamus to STN within the beta band. STN and striatum do not have all three electrode traces due to preprocessing steps. The striatum and thalamus depth recordings based off the top border, STN based off the middle of the structure, depths determined intraoperatively.
Peaks within the beta band
Figure 3: Peaks within the power spectrum (dB) beta band (12-40 Hz) for striatum, thalamus, and STN for one specific patient. Demonstrates how the findpeaks function locates the local maxima of a given signal and how those results were utilized to generate Table 1. The striatum and thalamus depth recordings based off the top border, STN based off the middle of the structure, depths determined intraoperatively.
14


Table 1
Min Max Mean Median Standard Deviation Coefficient of Variation K-S P- value A-B P- value F-test P- value
Striatum 1 6 2.8286 3 1.4448 0.51078 0.776 0.013 0.041
Thalamus 1 5 2.5366 2 1.1640 0.45888 0.549 0.082 0.341
STN 1 5 2.8889 3 0.9740 0.33716
In order to confirm that the findings were not specific to only certain depths along the trajectory, the information was visualized to show the change in power for small changes in trajectory depth for specific frequencies. A heatplot of the power spectrum for all depths recorded along the trajectory from 0-50 Hz is shown in Figure 4 to describe the increase in power noted from Figure 2 with respect to depth and intensity for one individual patient. From the color scale it is clear that from 0 to approximately 29 Hz the power is predominantly above 0 dB with high activity noted in the lower delta and theta frequency bands.
Heatplot through striatum, thalamus, and STN
Figure 4: Heatplot of power spectral density across all depths recorded in electrodes 2 and 3 within the 0-50Hz range. The vertical dotted lines indicate the start and end of the beta band, 12-40 Hz, the horizontal lines represent the top of the thalamus and the middle of the STN as determined intraoperatively. The power in dB is represented by color intensity with a maximum defined as 8 and minimum defined as -25 for the colorbar. The power visibly increases as
15


the electrode moves closer to the STN as shown by the dense region of higher power below the middle of STN for electrode 2 and slightly above the middle of STN for electrode 3.
Within the beta band the power gradually increases as the depth decreases with a more
notable intensity shift appearing between 3.167mm and 1.366mm. This increase in intensity is very
notable within the beta band between 22-29 Hz for this specific patient with the power approximately
5 dB. This figure supports the literature that there is an increase in power within the beta band for the
STN compared to other frequency bands and other structures.
Distinct Increase for Average Relative Power Emerges Across Multiple Patients
To determine if all patients showed similar trends across the major brain structures along the
trajectory the average power was calculated at the approximate depths for each structure along the trajectory for each electrode as seen in Figure 5. The depth numbers are with respect to the target, the ventral border of STN with striatum approximately 21-16 mm, thalamus 11-6 mm, and STN 6-1 mm. The power continuously increases as it moves down closer to the STN for majority of all patients as visible through the red lines. The average across all patients is shown using the black line and it
16


continuously increases.
Relative power along electrode trajectory
GO
T3
w
ro
Q)
L.
u
c
0)
5
o
a
u
>
H
4->
J2
a*
Figure 5: Average relative power increase across multiple patients within electrode 1 for approximate locations for the striatum (21-16 mm), thalamus (11-6 mm), and STN (6-lmm). The red lines represent individual patients, the black line represents the average of across all patients. All data was scaled to have the initial starting point at approximately zero by dividing by the minimum.
From preliminary analysis it is clear that the power increases continuously as the electrode nears the STN. To further validate these findings the brain structures of interest were identified along the trajectory from each surgery and those depths were used in the next figure to confirm that these findings were still applicable to more precise definitions of each structure.
Average Relative Power Continuously Increased from Striatum to Thalamus to STN
Focusing on specific depths recorded and known to be within each region, a comparison between the average relative power in the striatum, thalamus, and STN for all patients is seen
17


displayed as boxplots in Figure 6.
Boxplot of average relative power
Striatum Thalamus STN
Figure 6: Boxplot of the average relative power across multiple structures: striatum, thalamus, and STN. For striatum: 25* percentile = 48.07, median= 69.07, 75th percentile = 75.12, for thalamus: 25th percentile = 48.16, median = 68.25, 75th percentile = 80.83, for STN: 25th percentile = 49.84, median = 75.71, 75th percentile = 87.16. There is an increase in median and 75th percentile from striatum and thalamus to STN. The striatum and thalamus depth recordings based off the top border, STN based off the middle of the structure, depths determined intraoperatively. Relative power was calculated by dividing the signal by the maximum value for that specific depth recording.
The x-axis features the location along the trajectory, with striatum, thalamus, and STN in red, green, and blue respectively. The y-axis features the average relative power in percentage for each patient. The colored dots represent the individual electrode patient data points for each of the three structures. The box plot lines indicate the 25th ,75th percentile and range for each brain structure, the solid black line indicates the median or 50th percentile for each brain structure. The 25th percentile, median, and 75th percentile are higher for the STN compared to the striatum and thalamus. The 25th percentile is approximately the same from striatum to thalamus, 48.07% to 48.16%, but then increases for the STN to 49.84%. The median values for each brain structure increase from striatum and thalamus to STN with relative average power values of 69.07%, 68.25%, and 75.71%. Similar to the median values the 75th percentile values also increase through each structure by approximately 5-7%,
18


from 75.12% to 80.83% to 87.16% It is clear that nearly all patients have an upward trend between the
different structures with the highest average relative power for most being in the STN. These results were tested using a one-way ANOVA across structures within subjects and the resulting p-value was 0.6467, which indicates that the differences between the variables are not statistically significant. The one-way ANOVA results could be skewed due to the small sample size that was tested; if more data was tested it could be a smaller p-value indicating a more statistically significant difference between these three distinct brain structures, the striatum, thalamus, and STN
19


CHAPTER IV
DISCUSSION
This research sought to determine if there was an easily distinguishable way to differentiate known brain structures along the surgical trajectory in DBS using LFPs. The LFP signals were converted into the frequency domain and were analyzed, focusing specifically on the beta frequency band (12-40 Hz). By breaking the LFP signals into their power spectrum, it became clear that there were characterizable differences between the STN and other structures. Notably, there was an increase in power amplitude within the beta band, an increased number of oscillations within the power spectrum beta band, a clear separation of patient groups based on average relative power, and a continual increase in power within the beta band from the striatum to thalamus to STN.
One of the most significant findings was that the power spectrum amplitude increased for the STN compared to the other structures (striatum and thalamus). This finding supports current findings that have been published in recent literature. This increased amplitude was most notable within the beta frequency band of the LFPs for the PD patients. This finding was observed across multiple patients and was consistent not only at a specific depth identified to be the middle of STN (see Figure 2), but also followed along as distance to the STN decreased (see Figure 3). The increase in activity while distinct to the beta band could have been due to other outside sources instead of being a characteristic of the STN. Throughout the recordings, particularly in the STN, the surgeon is instructing the patient to complete various motor and non-motor tasks, so it is possible that increased motor activity could have skewed the power spectrum; however motor tests are also performed dorsal to the STN as well. Another limitation could be that these STN signals are defined as being in the middle of the structure, whereas the striatum and thalamus recordings are defined at the border, which could have resulted in a different signal than in the true center of those structures.
In addition to the increased amplitude for the power spectrum, the number of oscillations also increased when the electrodes were in the STN compared to the other brain structures along the
20


trajectory. This increased number of oscillations within that specific beta frequency band could be attributed to increased firing within the STN that manifests within the beta band range of frequencies. This is something that has not been noted within current literature but is a logical conclusion if the power amplitude is increased it is possible that it can be due to increased firing. Future studies will take advantage of the single unit and LFP data that are collected from the same depth location, at the tip of the microelectrode, to correlate spike frequency with LFP spectral magnitude, which will indicate whether changes in the local network influence population level activity in the STN. The preliminary statistical analyses indicate that within the STN and striatum data sets there are different distributions, which is very promising as far as identifying specific structures. Future directions will follow up on and clarity the differences within the peak oscillation data for the striatum, thalamus, and STN. One limitation of this method could be related to the LFP recordings being influenced by surrounding signals which can cause this appearance of oscillations without there actually being oscillations due to activity solely within the STN. Another major limitation could be that the sampling rate to convert the signal and import it into Matlab excluded signals that occurred in the higher frequency bands, which could have shown even smaller oscillations.
Once the average power was calculated for each patient for each approximate brain structure it was clear that the increase in power was visible for nearly all patients and not only limited to a few. These results are profound because of the larger number of patients this was observed in as a general trend, it was also present in all three electrodes. These results are also unique because all of the data recordings were from awake and conscious humans and thus subject to the high variable recordings associated with awake brain signals. During the recordings it is possible for the patient to talk, to move, and it is also possible for the electrodes to pick up on noise from other circuits firing in nearby regions, creating a highly variable and noisy recording to analyze.
When the average relative power of each signal was calculated for each specific structure the results showed a clear increase from striatum to thalamus to STN and SNr. Although an ANOVA did not indicate a main effect, there is a clear distinction between each of the structures that their relative
21


powers are in different ranges. The low p-value also could be related to the relatively sensitive ranges between the different structures as well as the small number of samples used (note that although the sample may not be sufficient for statistical significance it is a greater sample than normally examined in this experimental population). The increased power once normalized to be compared across all patients is consistent with the literature findings that power is increased for STN within the beta band. This could be due to increased firing or increased activity within the beta band. Some limitations of this include the limited number of samples, only depths indicated to be at precise locations within each structure were used to calculate the average relative power for each of the electrodes used, so if an electrode recording from the specified depth had any sort of artifact then it was not used. It is likely that an increased sample size would improve the sensitivity of the outcome.
A majority of these findings and analyses were focused on only the beta band due to the high focus in an LFP review from Thompson et al., 2014, although future studies will include analyzing the other known frequency bands and looking to see if the same trends are visible in those specific bands or if there are other more unique or more distinct ways to identify the borders of the STN from the surrounding structures. The LFP data is extremely usable and now that most preliminary processing and analyses have been performed focusing on one band, it will be very straightforward to begin to use the same approach to analyze different bands. It also could be interesting to look at other signal characteristics such as phase amplitude coupling and root mean square of the signals instead of mainly focusing on the power spectrum for each signal.
22


CHAPTER V
CONCLUSION
These findings indicate the LFPs are robust signals that are useful for determining different structures within the trajectory for DBS in PD patients. These signals give a broader implication of what is going on and can be more useful for localizing the STN. There were a variety of identifiers noted for the STN within the beta frequency range that could be incorporated into future DBS procedures to help localize the target quicker and with more accuracy. The development of a system with this technology integrated in could lead to a closed-loop device, which is more efficient and effective at treating PD motor symptoms.
23


BIBLIOGRAPHY
1 Fritsch, G., & Hitzig, E. (2009). Electric excitability of the cerebrum (Uber die elektrische Erregbarkeit des Grosshirns). Epilepsy & Behavior, 15(2), 123-130.
" Spiegel, E. A., Wycis, H. T., Marks, M., & Lee, A. J. (1947). Stereotaxic Apparatus for Operations on the Human Brain. Science, 106(2754), 349-350.
Sem-Jacobsen, C. W. (2009). Depth Electrographic Stimulation And Treatment Of Patients With Parkinson's Disease Including Neurosurgical Technique. Acta Neurologica Scandinavica,
41(S13), 365-376.
IV Cooper, I. S., Riklan, M., Amin, I., Waltz, J. M., & Cullinan, T. (1976). Chronic cerebellar stimulation in cerebral palsy. Neurology, 26(8), 744-744.
v Brice, J., & Mclellan, L. (1980). Suppression Of Intention Tremor By Contingent Deep-Brain Stimulation. The Lancet, 315(8180), 1221-1222.
Vl Benabid, A., Poliak, P., Hoffmann, D., Gervason, C., Hommel, M., Perret, J., Gao, D. (1991). Long-term suppression of tremor by chronic stimulation of the ventral intermediate thalamic nucleus. The Lancet, 337(8738), 403-406.
v" Perlmutter, J. S., & Mink, J. W. (2006). Deep Brain Stimulation. Annu. Rev. Neurosci. Annual Review of Neuroscience, 29(1), 229-257.
Vl" Williams, N. R., & Okun, M. S. (2013). Deep brain stimulation (DBS) at the interface of neurology and psychiatry. Journal of Clinical Investigation J. Clin. Invest., 123(11), 4546-4556.
IX Benabid, A. L., Poliak, P., Gao, D., Hoffmann, D., Limousin, P., Gay, E., Benazzouz, A. (1996). Chronic electrical stimulation of the ventralis intermedius nucleus of the thalamus as a treatment of movement disorders. Journal of Neurosurgery, 84(2), 203-214. x Anderson, V. C., Burchiel, K. J., Hogarth, P., Favre, J., & Hammerstad, J. P. (2005). Pallidal vs Subthalamic Nucleus Deep Brain Stimulation in Parkinson Disease. Arch Neurol Archives of Neurology, 62(4), 554.
Xl Diederich, N. J., Kalteis, K., Stamenkovic, M., Pieri, V., & Alesch, F. (2005). Efficient internal pallidal stimulation in Gilles de la Tourette syndrome: A case report. Movement Disorders Mov Disord., 20(11), 1496-1499.
x" Ranck, J. B. (1975). Which elements are excited in electrical stimulation of mammalian central nervous system: A review. Brain Research, 98(3), 417-440.
Xl" Recently-Approved Devices : Activa Parkinson's Control System P960009/S7. (n.d.).
Retrieved March 08, 2016, from
http://www.fda.gov/MedicalDevices/ProductsandMedicalProcedures/DeviceApprovalsandClearan
ces/Recently-ApprovedDevices/ucm083894.htm
XIV Rodriguez-Oroz, M. C. (2004). Efficacy of deep brain stimulation of the subthalamic nucleus in Parkinson's disease 4 years after surgery: Double blind and open label evaluation. Journal of Neurology, Neurosurgery & Psychiatry, 75(10), 1382-1385.
Mink, J. W. (1996). The Basal Ganglia: Focused Selection And Inhibition Of Competing Motor Programs. Progress in Neurobiology, 50(4), 381-425.
Smith, Y., Hazrati, L., & Parent, A. (1990). Efferent projections of the subthalamic nucleus in the squirrel monkey as studied by the PHA-L anterograde tracing method. The Journal of Comparative Neurology J. Comp. Neurol., 294(2), 306-323.
xv" Williams, N. R., Foote, K. D., & Okun, M. S. (2014). Subthalamic Nucleus Versus Globus Pallidus Internus Deep Brain Stimulation: Translating the Rematch Into Clinical Practice. Movement Disorders Clinical Practice Mov Disord Clin Pract, 1(1), 24-35.
XVI" Little, S., & Brown, P. (2012). Brain Stimulation in Neurology and Psychiatry. Annals of the New York Academy of Sciences, 1265(1), 9-24.
XIX Schwartz, A. B., Cui, X. T., Weber, D., & Moran, D. W. (2006). Brain-Controlled Interfaces: Movement Restoration with Neural Prosthetics. Neuron, 52(1), 205-220.
Bronte-Stewart, H., Barberini, C., Koop, M. M., Hill, B. C., Henderson, J. M., & Wingeier, B. (2009). The STN beta-band profile in Parkinson's disease is stationary and shows prolonged attenuation after deep brain stimulation. Experimental Neurology, 215(1), 20-28.
24


**' Foffani, G., Bianchi, A. M., Baselli, G., & Priori, A. (2005). Movement-related frequency modulation of beta oscillatory activity in the human subthalamic nucleus. The Journal of Physiology, 568(2), 699-711.
xx" Little, S., & Brown, P. (2012). Brain Stimulation in Neurology and Psychiatry. Annals of the New York Academy of Sciences, 1265(1), 9-24.
25


Full Text

PAGE 1

USING LOCAL FIELD POTENTIALS TO DIFFERENTIATE BRAIN REGIONS AND OPTIMIZE TARGET LOC ALIZATION FOR DEEP BRAIN STIMULATION SURGERY by RACHEL VERA KOLB B.S., The College of New Jersey, 2010 A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the d egree of Masters of Science Bioengineering Program 2016 08 !"## $

PAGE 2

ii This thesis for the Master of Science degree by Rachel Vera Kolb has been approved for the Bioengineering program by Emily Gibson Chair Gidon Felsen John Thompson April 28 th 2016

PAGE 3

iii Kolb, Rachel Vera MS, Bioengineering Using L ocal Field Potentials to Differentiate Brain Regions and Optimize Target Localization for Deep Brain Stimulation Surgery Thesis directed by Assistant Professor Emily Gibson ABSTRACT Deep Brain Stimulation (DBS) is a common surgical treatment option for in dividuals with movement disorders like Parkinson's disease (PD), and requires precise electrode placement to electrically stimulate the target brain structure. Traditionally for DBS, electrode placement is determined based on non invasive imaging technique s and the stereotactic coordinates are computed using software. To aid in more accurate DBS electrode placement than simply the computer generated location, I examined whether local field potentials (LFPs) obtained from the macroelectrode contacts could di stinguish when the electrode is within the target structure because local field potentials give a more global recording perspective than single unit recordings Several studies have indicated that increased power spectral activity in the beta band (12 40 H z) in the STN has been associated with akinetic motor output in PD. To compare these results using LFPs, the power spectrum was analyzed in Matlab using standard signal processing techniques I t became clear that within the beta band there are signatures t hat distinctly separate the target STN from other structures along the trajectory. The LFP power spectrum showed increased amplitude and an increased number of oscillations in the beta band for depths confirmed to be in the middle of the STN. Analysis of t he average relative power for specific depths identified along the trajectory indicated that power continuously increases from striatum to thalamus to STN. These features of LFPs within the beta band seem to be distinct markers of the STN and could be inco rporated into future technologies to more confidently identify the target electrode location and validate the use of LFPs to monitor during DBS The form and content of this abstract are approved. I recommend its publication. Approved: Emily Gibson

PAGE 4

iv TABLE OF CONTENTS CHAPTER I INTRO """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""" # What is Deep Brain Stimulation? """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""" # DBS and Parkinson's Disease """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """"""""""""""""""""""""""""" $ Using Local Field Potentials to study DBS for PD """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """"""""""""""""""""" % II METHODS """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """"""""""""""" & DBS Surgery """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """"""""""""""""""""""""""""" & LFP Electrophysiology """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""" & Subjects & Sample Size """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""" Matlab Analysis """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """"""""""""""""""""""" III RESULTS """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""" #( Increase in Amplitude and Number of Signal Oscillations for the Power Spectral Density of the STN Within the Beta Band """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """"""""""""""""""""""""""""""" #( Distinct Increase for Aver age Relative Power Emerges Across Multiple Patients """""""""""""""""""""""" #) Average Relative Power Continuously Increased from Striatum to Thalamus to STN """"""""""""""" #* IV DISCUSSION """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""" $( V CONCLUSI ON """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """"" $% BIBLIOGRAPHY """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """""""""""""""""""""""""""""""" """"" $&

PAGE 5

1 CHAPTER I INTRO What is Deep Brain Stimulation? The development of deep brain stimulation ( DBS ) as a therapeutic tool has been evolving since 1870 when Fritsch & Hitzi ng demonstrated localized electrical stimulation can be used to investigate brain function i A lthough it wasn't until the development of stereotaxic devices closer to 1957 that deeper stimulation of the brain began to be explored. ii In the 1960s the concept of high frequency stimulation to treat movement disorders became popular and surgical lesions were targeted by implanting multiple electrodes together using wires and stimulating them for weeks at a time. iii DBS continued to be used for epilepsy, cerebral p alsy and Parkinson's Disease (PD) treatment, and throughout the 1970s iv and 1980s v It wasn't until the 1990s when DBS and chronic stimulation technology from pacemakers became more prevalently accepted as a treatment option for a variety of disorders. vi To day, DBS is considered a reversible neurosurgical treatment that can be used for advanced movement disorders such as: dystonia, essential tremor, and PD. vii It is also an emerging therapy that has the potential to provide relief for treatment resistant dis eases such as obsessive compulsive disorder, Tourette's, and other neuropsychiatric conditions. viii Different sites of stimulation within deep brain structures have been linked to ameliorating specific symptoms : essential tremor has been linked to the ventral intermediate thalamus ix PD and dystonia have been linked to both the Globus pallidus interna (GPi) and subthalamic nucleus (STN) x and recently Tourette syndrome has been linked to the GPi. xi DBS surgery involves the stereotactic implantation of a quadrip olar macroelectrode into a deep brain structure to disrupt the abnormal neuronal firing rates/patterns to treat motor symptoms. DBS can have varying effects throughout structures in the brain due to characteristics like stimulation amplitude, temporal spac ing, and properties of individual cells, stimulus field geometry, target geometry, and even the underlying pathophysiology of the disease state. xii The Federal Drug Administration of the US approved the first DBS device, Medtronic's Activa DBS Therapy System for essential tremor in 1997 and PD in 2002 xiii

PAGE 6

2 DBS and Parkinson's Disease The motor symptoms of PD can be managed using medication which increases or substitutes dopamine within the brain, commonly levodopa is used because it is a natural chemical whic h is converted into dopamine once it passes into the brain. Over time, levodopa based treatments become less effective at managing movement disorders, requiring higher dosages and can ultimately lead to dyskinesia. Patients with PD that are no longer respo nsive to pharmacological and nonpharmacological treatments can consider DBS as the next option to manage their disease symptoms. For PD DBS targets can be either the STN or GPi to optimize patient outcome but the most common target is the STN a node wit hin the basal ganglia circuit. xiv Within the basal ganglia circuit, the STN receives excitatory inputs from the cerebral cortex, other basal ganglia components, thalamus pedunculopontine nucleus and substantia nigra pars compacta and inhibitory inputs from Globus pallidus external. xv The STN has glutamatergic, excitatory outputs to Globus pallidus, substantia nigra pars reticulata ( SNr ) and pedunculopontine area. xvi Target ing the STN has also been associated with reduced medication and less frequent battery c hanges because the target area is smaller, resulting in a better long term economic profile xvii The STN target is initially located using preoperative non invasive imaging and then informed using intraoperative electrophysiological techniques. Current surgic al practice depends upon single unit electrophysiology to define the optimal implant target. Single unit potentials are highly accurate and taken from a small sample of neurons (6 10). They are highly susceptible to mechanical and electrical perturbation d uring recordings, greatly reducing their reliability It is possible to collect local field potentials (LFPs) during DBS along the trajectory to the target from both the microelectrode and the DBS electrode without requiring any additional electrodes or ha rdware LFPs represent the population averaged signal of electrical current flowing in nearby neurons and give a larger sample of neurons ( ~ 1000). xviii These LFP signals can be recorded at a lower frequency than single unit recordings which results in signals less affected by electrode interface or local geometry. xix Electrical signal recordings obtained from DBS can be separated into frequency bands and then associated with relevant clinical signatures: delta (0 4 Hz), theta (4 8 Hz), alpha (8 12 Hz), beta (12 40 Hz), and gamma (40 100 Hz). LFP recordings in DBS

PAGE 7

3 with a n STN target from LastName et al., ( 2009 ) indicated that there was elevated LFP power within the beta frequency band linked to motor symptoms of PD, like rigidity and bradykinesia xx Frequency modu lation in the beta band is regulated by circulating dopamine levels which is consistent with PD because it is caused by the death of dopaminergic neurons. xxi The STN is where information processing information from across the cortex converges so the populat ion based metric of LFP is a more logical choice than the single unit recordings to look at broader state changes in PD. xxii Using Local Field Potentials to study DBS for PD If it is possible to identify distinct signal characteristics for specific regions along the electrode trajectory LFPs could be utilized for determining the borders of different brain structures. The LFP signatures within the beta band of the STN could also be utilized as a biomarker to develop a closed loop stimulation DBS device. LFPs are highly robust signal s that represent an entire neuronal population and have the potential to aid in surgical targeting of specific brain regions during DBS. In this study, we examined whether LFP signals could be used to objectively predict entry and exit of STN in PD patients undergoing DBS. Distinct characteristics emerged within the beta band of the recorded LFPs to differentiate the STN from other structures along the surgical trajectory, including an increase in instantaneous power, increased numb er of peaks, and increased relative average power. This study confirmed that LFP can be utilized as a tool to enhance DBS and improve target localization for the STN.

PAGE 8

4 CHAPTER II METHODS All procedures, data collection and analyses were approved by the Colorado Multiple Institution Review Board (COMIRB). All LFP analyses were conducted on de identified data from patients who provided their voluntary and informed consent. DBS S urgery Data were collected during routine DBS implantation surgery (describe d in detail Abosch et al., 2013 (Stereotactic Functional Neurosurgery 91(1):1 11) ). Briefly, pre operative brain scans both magnetic resonance (MR) and computed tomography (CT) with a contrast enhancing head frame are imported into stereotactic plannin g software to derive initial STN target coordinates associated with a CRW (Cosman Robert Wells) stereotactic system The stereotactic frame is aligned on the patient based on the planned trajectory from the computer software Three microelectrodes are used to traverse the planned trajectory starting at 25 mm above target (ventral border of STN) and are spaced 2mm apart from each other. The microelectrodes are advanced in submillimetric steps (100 500 !m). High impedance recordings from the microelectrodes a re displayed on an oscilloscope and used to visually assess location in the brain and ultimately verify the DBS target. Once the STN is identified, patients engage in arm and leg motions to drive activity in the STN associated with somatotopic territory. D uring microelectrode recordings, patients are awake and initiating arm and leg movements Once the microelectrodes are determined to be in the STN, a single lead is chosen to implant Following macrostimulation to assess therapeutic benefit and side effect p rofile, the DBS electrode is secured in place LFP E lectrophysiology The microelectrode on each lead and the DBS macroelectrode are capable of recording LFPs (1.3 kHz sampling frequency) The LFP recordings are taken in incremental steps (100 500 !m) alon g the trajectory for at least 10 seconds for each signal recording. The LFP signals are digitally sampled and exported using specialized software to Matlab

PAGE 9

5 Subjects & Sample Size The obtained LFPs are from patients that underwent DBS surgery between 201 4 2016 at University of Colorado Hospital. All subjects have been diagnosed with PD and the target implantation for DBS surgery's was the STN. A total of 29 LFP recordings were analyzed, with 23 used for the analysis. All raw data was imported into Matlab and viewed to check for any recording issues or outliers. Some patient files were not used for analysis due to abnormalities in the signal recordings stemming from noise and interference. Matlab Analysis All of the LFP recordings were sampled at 1.3 k H z prior to being imported into Matlab The patient re cordings were run through a pre processing code written in Matlab to remove the mean from each electrode and band pass the signals using a 4 th order zero phase lag filter at 350 Hz Any data above 21mm fr om the target was removed due to proximity to an impedance check that could corrupt the data. The mean of the data was removed by evaluating each CLFP electrode separately for each depth data file, using the nan mean' function in Matlab to find the average value of an array ignoring any nan values which was subtracted from the overall signal for each electrode at each depth. A digital filter was used to make sure that the Nyquist frequency rule was met, that the minimum sampling rate of the signal is twice the highest frequency present to prevent any introduction of errors. If the Nyquist frequency was above 350 Hz it was automatically set to 350 Hz for the passband filter. The designfilt' function was used to design a low pass infinite impulse response (II R) filter with a 4 th order and pass band frequency of 350 Hz based on the sampling frequency of the recording ~1500 Hz. The IIR is a commonly used digital filter that uses feedback to generate an infinite number of values for a single impulse An IIR filte r uses less memory and requires less calculations than other digital filters, a 4 th order filter was used based on desired specifications from the passband, stopband attenuation and transition widths. The data was indexed after the filtering to remove any recording depths preceding the impedance check. The resulting files were saved into a Matlab structure and saved to the local drive. The preprocessed data was then sent through a second preprocessing code written in Matlab designed to eliminate any depth r ecordings that have values more than six times the standard deviation

PAGE 10

6 above the mean. This was done by removing any nan values in the depths resulting from removal in the first preprocessing round. For each recording depth the median of the signal was foun d using the median' function and the interquartile range of the signal found by the iqr' function. A threshold was calculated by adding the resulting median value with six times the interquartile range. The absolute value of the signal, obtained using th e abs' function, was compared to that threshold to determine if there was a recording error then that recording depth was not used. The resulting signals after this check were again saved as a structure to the local drive. Once both preprocessing steps we re completed, the resulting data was analyzed. Prior to generating any figures all of the nan values were removed and the signal was multiplied by a bit resolution of 38.14 and then divided by 1000. For each depth, the first few moments of the recording n umber of points as half of the sampling frequency, were eliminated to ensure that there was no residual noise from the movement of the electrode down the trajectory. The time axis was calculated for each depth recording based on the number of points in the data and the sampling frequency. All recordings were originally recorded for at least 10 seconds in the operating room, but due to preprocessing slightly decreased. To calculate the power spectral density (PSD) of the signal, the pwelch' function from Ma tlab was used with a moving average filter using a window of 650 points and an overlap of 50 points. This function was chosen to calculate Welch's power spectral density estimate using an averaging estimator and because their power' input scales each esti mate of the PSD by the equivalent noise bandwidth of the signal. The resulting frequency output for the PSD calculation was then indexed to only include frequencies from 1 50Hz. That same index was used on the PSD output to only obtain PSD values within th e 1 50Hz range. That resulting signal was converted into decibels by multiplying it by 10 log of base 10. To ensure that the recorded signals were all distinct for each depth, a magnitude squared coherence test was performed using the mscohere' function, it finds the magnitude squared coherence e stimate using Welch's

PAGE 11

7 average modified periodogram method. Figure 1 : (A) Schematic of surgical procedure set up and trajectory (B) Sample of time spectrum recording of three raw LFP sig nals. (C) Sample of power spectrum for three LFP signals including their standard deviation. (D) Sample signal coherence between all three LFP signal recordings indicating that each signal is distinct The data was entered into the function along with a 1 00 point hamming window with an overlap of 80, for frequencies under 100 at the given sampling frequency of ~1500Hz To find the mean for one single depth across all electrodes the nanmean' function was again used for all three electrode means to obtain a single mean value for that depth. To calculate the maximum value for one specific depth the max' function was used for all three electrodes to find a single maximum value at that depth. To calculate the mean power spectral density for one patient across all depths the nanmean' function was used again to find the average ignoring any nan values for each electrode individually

PAGE 12

8 and then used again on an array with each electrode's mean across the entire frequency to calculate the average power for frequenci es 1 50Hz across all depths to use as a reference line. To calculate the number of peaks within specific bands, the frequency band of interest was indexed into the calculated power signal for that specific depth. That resulting power signal for the frequen cy band of interest was used as an input into the findpeaks' function which finds the values of the local maxima. The length of the resulting array was calculated using the length' function to count the number of peaks. To analyze the distributions and variances of the number of peaks for each structure three preliminary statistical tests were done using functions within Matlab The two sample Kolmogorov Smirnov test to determine if both samples are from the same continuous distribution using the kstest 2' function, the Ansari Bradley test to determine if both samples are from the same distribution using the ansaribradley' function, and a two sample F test to determine if both samples have normal distributions with the same variance using the vartest2' function. To generate the heatplot showing power across all patient depths from 1 50Hz, power was calculated the same way as above. The power data was then sorted by depth using the sort' function in Matlab to order the depths in descending order. Then an y data rows that began with a nan value were removed using ~isnan' logic, to only leave values that were not a nan. To display the image, imagesc' was used to visualize it with scaled colors, with an image property setting of CDataMapping' to scaled' t o allow us to define the minimum and maximum color limits for the image to standardize across all electrodes as 25 to 8 dB. To compare multiple patients together and look at the power spectrum trends for the striatum thalamus and STN within the beta ban d each structure depth was approximated and the average power calculated. The average power was calculated the same way as described above for the following depth ranges, striatum 21 to 16 mm, space between striatum and thalamus 16 to 11 mm, thalamus 11 to 6 mm, STN 6 to 1 mm, and SNr 1 to 4mm. The average power for each region was calculated separately for each electrode and then all patients for that electrode were plotted together using the patchline' function, then the average for all patients for eac h electrode was calculated by taking the mean and plotting it using the same patchline' function. To precisely identify the structures for each patient specific to each electrode an excel file was created that included the specific depths

PAGE 13

9 that each struct ure was identified to be in from each surgery. The values found were defined to be the middle of the striatum the top of the thalamus the middle of the STN, and where applicable, the top of the SNr For each structure the power was calculated by electro de for each individual the same way as outlined above. In order to compare across all patients the average relative power was calculated by dividing the power for each specific structure by the maximum power found for that individual within that electrode. The median for each structure was calculated across all patients using the nanmedian' function to ignore nan values for missing data. The 25 th and 75 th percentiles were calculated using the prctile' function for the data. An ANOVA was run on the data to compare the striatum thalamus and STN using the anova1' function. The SNr was not included in the ANOVA because there were not enough confirmed points within that structure.

PAGE 14

10 CHAPTER III R ESULTS Increase in A mplitude and N umber of S ignal O scillatio ns for the P ower S pectral D ensity of the STN W ithin the B eta B and Current literature has shown significant increases in the power spectral density amplitude within the STN for the beta band. In order to evaluate these previously published findings with our datasets, the multitaper power spectral density was calculated. T he LFP power spectral density for four of the brain structures confirmed along the trajectory: striatum thalamus and STN is depicted in Figure 2 Each graph in F igure 2 includes LFP elec trode recordings and the amplitude mean and maximum for that specific depth recording. It is clear that for the beta frequency band, 12 40 Hz, the power increases from striatum to thalamus and thalamus to STN. For the thalamus recording the LFP electrode traces are very close to the overall mean dashed line trace, and they clearly shift above that same line for the STN recording. The mean of the signal for the STN is higher for STN than for thalamus, 5.25 vs 4.93 dB. In F igure 3 between the thalamus and STN there also is a clear increase in oscillatory activity, specifically between 30 35 Hz. The thalamus has 2 local maxima whereas the STN has 5 local maxima. This figure supports the literature findings that the number of oscillations increase in the beta band for the STN compared the structures found higher along the trajectory like the striatum and thalamus. The analysis for the number of peaks and the statistical significance of the various tests run on the data to compare the striatum and thalamus to the STN can be seen in Table 1 The two sample Kolmogorov Smirnov test was used to determine if both samples are from the same continuous distribution, the Ansari Bradley test was used to determine if both samples are from the same distribution, and a two sample F test was used to determine if both samples have normal distributions with the same variance. The standard deviation for the STN is much lower than the striatum or thalamus and the mean and median is higher, indicating that there is less variation and larger values. The STN also has the smallest coefficient of variation for the three structures. The p values found comparing the striatum and thalamus to the STN indicate that there are statistically

PAGE 15

1 1 significant differences in the distribution between the STN and the striatum and that the values are not from the same distributions.

PAGE 16

12

PAGE 17

13

PAGE 18

14 Figure 2 : LFP Multitaper power spectral density in decibels for 12 4 0 Hz within the striatum thalamus and STN Striatum recording has a mea n of 1.93 dB, maximum of 9.02 dB, thalamus recording has a mean of 2.63 dB, maximum of 11.13 dB, STN has a mean of 4.80 dB, maximum of 10.16 dB. The mean increases consistently from striatum to thalamus to STN within the beta band. STN and striatum do not ha ve all three electrode traces due to preprocessing steps. The striatum and thalamus depth recordings based off the top border, STN based off the middle of the structure depths determined intraoperatively. Figure 3 : Peaks within the power spectrum (dB) beta band (12 40 Hz) for striatum, thalamus, and STN for one specific patient. Demonstrates how the findpeaks' function locates the local maxima of a given signal and how those results were utilized to generate Table 1. The striat um and thalamus depth recordings based off the top border, STN based off the middle of the structure, depths determined intraoperatively.

PAGE 19

15 Table 1 Min Max Mean Median Standard Deviation Coefficient of Variation K S p value A B p value F test p value Striatum 1 6 2.8286 3 1.4448 0.51078 0.776 0.013 0.041 Thalamus 1 5 2.5366 2 1.1640 0.45888 0.549 0.082 0.341 STN 1 5 2.8889 3 0.9740 0.33716 ---In order to confirm that the findings were not specific to only certain depths along the trajectory, the information was visualized to show the change in power for small changes in trajectory depth for specific frequencies. A heatplot of the power spectrum for all depths recorded along the trajectory from 0 50 Hz is shown in Figure 4 to describe the increase in power noted from Figure 2 with respect to depth and intensity for one individual patient From the color scale it is clear that from 0 to approximately 29 Hz the power is predominantly above 0 dB with high activity noted in the lower delta and theta frequency bands. Figure 4 : Heatplot of power spectral density across all depths recorded in electrodes 2 and 3 within the 0 50Hz range The vertical dotted lines indicate the start and end of the beta band, 12 40 Hz, the horizontal lines represent the top of the thalamus and the middle of the STN as determined intraoperatively. The power in dB is represented by color intensity with a maximum defined as 8 and minimum defined as 25 for the colorbar. The power visibly increases as

PAGE 20

16 the electrode moves closer to the STN as shown by the dense region of higher power below the middle of STN for electrode 2 and slightly above the middle of STN for electrode 3. Within the beta band the power gradually increases as th e depth decreases with a more notable intensity shift appearing between 3.167mm and 1.366mm. This increase in intensity is very notable within the beta band between 22 29 Hz for this specific patient with the power approximately 5 dB. This figure supports the literature that there is an increase in power within the beta band for the STN compared to other frequency bands and other structures Distinct I ncrease for A verage R elative P ower E merge s A cross M ultiple P atients To determine if all patients sh owed similar trends across the major brain structures along the trajectory the average power was calculated at the approximate depths for each structure along the trajectory for each electrode as seen in Figure 5 The depth numbers are with respect to the target the ventral border of STN with striatum approximately 21 16 mm, thalamus 11 6 mm, and STN 6 1 mm The power continuously increase s as it moves down closer to the STN for majority of all patients as visible through the red lines. The average across all patients is shown using the black line and it

PAGE 21

17 continuously increases. Figure 5 : Average relative power increase across multiple patients within electrode 1 for approximate locations for the striatum (21 16 mm) thalamus (11 6 mm) and STN (6 1mm) The red lines represent individual patients, the black line represents the average of across all patients. All data was scaled to have the initial starting point at approximately zero by dividing by the minimum. From preliminary ana lysis it is clear that the power increases continuously as the electrode nears the STN. To further validate these findings the brain structures of interest were identified along the trajectory from each surgery and those depths were used in the next figure to confirm that these findings were still applicable to more precise definitions of each structure. Average R elative P ower C ontinuously I ncreased from S triatum to T halamus to STN Focusing on specific depths recorded and known to be within each region, a comparison between the average relative power in the striatum thalamus and STN for all patients is seen

PAGE 22

18 displayed as boxplots in F igure 6 Figure 6 : Boxplot of the a verage relative power across multiple structures : striatum t halamus and STN. For striatum: 25 th percentile = 48.07, median= 69.07, 75 th percentile = 75.12, for thalamus: 25 th percentile = 48.16, median = 68.25, 75 th percentile = 80.83, for STN: 25 th percentile = 49.84, median = 75.71, 75 th percentile = 87.16. Ther e is an increase in median and 75 th percentile from striatum and thalamus to STN. The striatum and thalamus depth recordings based off the top border, STN based off the middle of the structure, depths determined intraoperatively. Relative power was calcula ted by dividing the signal by the maximum value for that specific depth recording. The x axis features the location along the trajectory, with striatum thalamus and STN in red, green, and blue respectively. The y axis features the average relative power in percentage for each patient. The colored dots represent the individual electrode patient data points for each of the three structures. The box plot lines indicate the 25 th 75 th percentile and range for each brain structure, the solid black line indica tes the median or 50 th percentile for each brain structure. The 25 th percentile, median, and 75 th percentile are higher for the STN compared to the striatum and thalamus The 25 th percentile is approximately the same from striatum to thalamus 48.07% to 48 .16 %, but then increases for the STN to 49.84 %. The median values for each brain structure increase from striatum and thalamus to STN with relative average power values of 69.07 %, 68.2 5%, and 75.71 %. Similar to the median values the 75 th percentile values also increase through e ach structure by approximately 5 7%,

PAGE 23

19 from 75.12% to 80.83% to 87.16 % It is clear that nearly all patients have an upward trend between the different structures with the highest average relative power for most being in the STN. These results were tested using a one way ANOVA across structures within subjects and the resulting p value was 0.6467 which indicates that the differences between the variables are not statistically significant. The one way ANOVA results could be skewed due t o the small sample size that was tested; if more data was tested it could be a smaller p value indicating a more statistically significant difference between these three distinct brain structures, the striatum thalamus and STN

PAGE 24

20 CHAPTER IV DISCUSSION T his research sought to determine if there was an easily distinguishable way to differentiate known brain structures along the surgical trajectory in DBS using LFPs. The LFP signals were converted into the frequency domain and were analyzed, focusing specif ically on the beta frequency band (12 40 Hz). By breaking the LFP signals into their power spectrum, it became clear that there were characterizable differences between the STN and other structures. Notably, there was an increase in power amplitude within the beta band, an increase d number of oscillations within the power spectrum beta band, a clear separation of patient groups based on average relative power, and a continual increase in power within the beta band from the striatum to thalamus to STN. On e of the most significant findings was that the power spectrum amplitude increased for the STN compared to the other structures ( striatum and thalamus ). This finding supports current findings that have been published in recent literature. This increased am plitude was most notable within the beta frequency band of the LFPs for the PD patients. This finding was observed across multiple patients and was consistent not only at a specific depth identified to be the middle of STN (see Fig ure 2 ) but also followe d along as distance to the STN decreased (see Fig ure 3 ) The increase in activity while distinct to the beta band could have been due to other outside sources instead of being a characteristic of the STN Throughout the recordings particularly in the STN the surgeon is instructing the patient to complete various motor and non motor tasks, so it is possible that increased motor activity could have skewed the power spectrum ; however motor tests are also performed dorsal to the STN as well Another limitat ion could be that these STN signals are defined as being in the middle of the structure, whereas the striatum and thalamus recordings are defined at the border, which could have resulted in a different signal than in the true center of those structures. I n addition to the increased amplitude for the power spectrum, the number of oscillations also increased when the electrodes were in the STN compared to the other brain structures along the

PAGE 25

21 trajectory. This increased number of oscillations within that speci fic beta frequency band could be attributed to increased firing within the STN that manifests within the beta band range of frequencies. This is something that has not been noted within current literature but is a logical conclusion if the power amplitude is increased it is possible that it can be due to increased firing. Future studies will take advantage of the single unit and LFP data that are collected from the same depth location, at the tip of the microelectrode, to correlate spike frequency with LFP spectral magnitude, which will indicate whether changes in the local network influence population level activity in the STN. The preliminary statistical analyses indicate that within the STN and striatum data sets there are different distributions, which i s very promising as far as identifying specific structures. Future directions will follow up on and clarify the differences within the peak oscillation data for the striatum thalamus and STN. One limitation of this method could be related to the LFP reco rdings being influenced by surrounding signals which can cause this appearance of oscillations without there actually being oscillations due to activity solely within the STN. Another major limitation could be that the sampling rate to convert the signal a nd import it into Matlab excluded signals that occurred in the higher frequency bands which could have shown even smaller oscillations. Once the average power was calculated for each patient for each approximate brain structure it was clear that the inc rease in power was visible for nearly all patients and not only limited to a few. These results are profound because of the larger number of patients this was observed in as a general trend, it was also present in all three electrodes. These results are al so unique because all of the data recordings were from awake and conscious humans and thus subject to the high variable recordings associated with awake brain signals During the recordings it is possible for the patient to talk, to move, and it is also po ssible for the electrodes to pick up on noise from other circuits firing in nearby regions, creating a highly variable and noisy recording to analyze. When the average relative power of each signal was calculated for each specific structure the results sh owed a clear increase from striatum to thalamus to STN and SNr Although an ANOVA did not indicate a main effect there is a clear distinction between each of the structures that their relative

PAGE 26

22 powers are in different ranges. The low p value also could be related to the relatively sensitive ranges between the different structures as well as the small number of samples used (note that although the sample may not be sufficient for statistical significance it is a greater sample than normally examined in this experimental population) The increased power once normalized to be compared across all patients is consistent with the literature findings that power is increased for STN within the beta band. This could be due to increased firing or increased activity wi thin the beta band. Some limitations of this include the limited number of samples, only depths indicated to be at precise locations within each structure were used to calculate the average relative power for each of the electrodes used, so if an electrode recording from the specified depth had any sort of artifact then it was not used. It is likely that an increased sample size would improve the sensitivity of the outcome. A m ajority of these findings and analyses were focused on only the beta band due to the high focus in an LFP review from Thompson et al., 2014 although future studies will include analyzing the other known frequency bands and looking to see if the same trends are visible in those specific bands or if there are other more unique or mor e distinct ways to identify the borders of the STN from the surrounding structures. The LFP data is extremely usable and now that most preliminary processing and analyses have been performed focusing on one band, it will be very straightforward to begin to use the same approach to analyze different bands. It also could be interesting to look at other signal characteristics such as phase amplitude coupling and root mean square of the signals instead of mainly focusing on the power spectrum for each signal.

PAGE 27

23 CHAPTER V C ONCLUSION These findings indicate the LFPs are robust signals that are useful for determining different structures within the trajectory for DBS in PD patients. These signals give a broader implication of what is going on and can be more use ful for localizing the STN. There were a variety of identifiers noted for the STN within the beta frequency range that could be incorporated into future DBS procedures to help localize the target quicker and with more accuracy. The development of a system with this technology integrated in could lead to a closed loop device which is more efficient and effective at treating PD motor symptoms.

PAGE 28

24 BIBLIOGRAPHY i Fritsch, G., & H itzig, E. (2009). Electric excitability of the cerebrum (†ber die elektrische Erregbarkeit des Grosshirns). Epilepsy & Behavior, 15(2), 123 130. ii Spiegel, E. A., Wycis, H. T., Marks, M., & Lee, A. J. (1947). Stereotaxic Apparatus for Operations on the Hum an Brain. Science, 106(2754), 349 350. iii Sem Jacobsen, C. W. (2009). Depth Electrographic Stimulation And Treatment Of Patients With Parkinson's Disease Including Neurosurgical Technique. Acta Neurologica Scandinavica, 41(S13), 365 376. iv Cooper, I. S., Ri klan, M., Amin, I., Waltz, J. M., & Cullinan, T. (1976). Chronic cerebellar stimulation in cerebral palsy. Neurology, 26(8), 744 744. v Brice, J., & Mclellan, L. (1980). Suppression Of Intention Tremor By Contingent Deep Brain Stimulation. The Lancet, 315( 8180), 1221 1222. vi Benabid, A., Pollak, P., Hoffmann, D., Gervason, C., Hommel, M., Perret, J., Gao, D. (1991). Long term suppression of tremor by chronic stimulation of the ventral intermediate thalamic nucleus. The Lancet, 337(8738), 403 406. vii Pe rlmutter, J. S., & Mink, J. W. (2006). Deep Brain Stimulation. Annu. Rev. Neurosci. Annual Review of Neuroscience, 29(1), 229 257. viii Williams, N. R., & Okun, M. S. (2013). Deep brain stimulation (DBS) at the interface of neurology and psychiatry. Journal o f Clinical Investigation J. Clin. Invest., 123(11), 4546 4556. ix Benabid, A. L., Pollak, P., Gao, D., Hoffmann, D., Limousin, P., Gay, E., Benazzouz, A. (1996). Chronic electrical stimulation of the ventralis intermedius nucleus of the thalamus as a treatment of movement disorders. Journal of Neurosurgery, 84(2), 203 214. x Anderson, V. C., Burchiel, K. J., Hogarth, P., Favre, J., & Hammerstad, J. P. (2005). Pallidal vs Subthalamic Nucleus Deep Brain Stimulation in Parkinson Disease. Arch Neurol Archi ves of Neurology, 62(4), 554. xi Diederich, N. J., Kalteis, K., Stamenkovic, M., Pieri, V., & Alesch, F. (2005). Efficient internal pallidal stimulation in Gilles de la Tourette syndrome: A case report. Movement Disorders Mov Disord., 20(11), 1496 1499. xii R anck, J. B. (1975). Which elements are excited in electrical stimulation of mammalian central nervous system: A review. Brain Research, 98(3), 417 440. xiii Recently Approved Devices : Activa¨ Parkinson's Control System P960009/S7. (n.d.). Retrieved March 08, 2016, from http://www.fda.gov/MedicalDevices/ProductsandMedicalProcedures/DeviceApprovalsandClearan ces/Recently ApprovedDevices/ucm083894.htm xiv Rodriguez Oroz, M. C. (2004). Efficacy of deep brain stimulation of the subthalamic nucleus in Parkinson's dise ase 4 years after surgery: Double blind and open label evaluation. Journal of Neurology, Neurosurgery & Psychiatry, 75(10), 1382 1385. xv Mink, J. W. (1996). The Basal Ganglia: Focused Selection And Inhibition Of Competing Motor Programs. Progress in Neurob iology, 50(4), 381 425. xvi Smith, Y., Hazrati, L., & Parent, A. (1990). Efferent projections of the subthalamic nucleus in the squirrel monkey as studied by the PHA L anterograde tracing method. The Journal of Comparative Neurology J. Comp. Neurol., 294(2), 306 323. xvii Williams, N. R., Foote, K. D., & Okun, M. S. (2014). Subthalamic Nucleus Versus Globus Pallidus Internus Deep Brain Stimulation: Translating the Rematch Into Clinical Practice. Movement Disorders Clinical Practice Mov Disord Clin Pract, 1(1), 2 4 35. xviii Little, S., & Brown, P. (2012). Brain Stimulation in Neurology and Psychiatry. Annals of the New York Academy of Sciences, 1265(1), 9 24. xix Schwartz, A. B., Cui, X. T., Weber, D., & Moran, D. W. (2006). Brain Controlled Interfaces: Movement Restor ation with Neural Prosthetics. Neuron, 52(1), 205 220. xx Bronte Stewart, H., Barberini, C., Koop, M. M., Hill, B. C., Henderson, J. M., & Wingeier, B. (2009). The STN beta band profile in Parkinson's disease is stationary and shows prolonged attenuation af ter deep brain stimulation. Experimental Neurology, 215(1), 20 28.

PAGE 29

25 xxi Foffani, G., Bianchi, A. M., Baselli, G., & Priori, A. (2005). Movement related frequency modulation of beta oscillatory activity in the human subthalamic nucleus. The Journal of Physiolo gy, 568(2), 699 711. xxii Little, S., & Brown, P. (2012). Brain Stimulation in Neurology and Psychiatry. Annals of the New York Academy of Sciences, 1265(1), 9 24.