| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
INNOVATIVE METHODOLOGY
1McGovern Institute for Brain Research and Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology; and 2The Rowland Institute at Harvard, Harvard University, Cambridge, Massachusetts
Submitted 12 June 2008; accepted in final form 15 September 2008
 |
ABSTRACT |
|---|
|
|
|
INTRODUCTION |
|---|
|
However, many brain structures of interest (e.g., subcortical structures, ventral cortex) are located deep within the brain and cannot be accessed directly from the skull surface. Performing fine-grained mapping of such structures is problematic because electrodes must travel large distances to reach their target and even small errors in alignment at the electrode insertion point can produce large position errors at the depth. For instance, at a depth of 40 mm (typical for recordings from ventral inferior temporal cortex of nonhuman primates), just a 2° angular error at the surface will produce a >1-mm displacement in the target brain structure. Such effects are compounded by the desire to use thinner microelectrodes (to minimize tissue damage), which are inherently more flexible, and thus less guaranteed to travel straight. In general, given these factors, "dead reckoning" of electrode target position from insertion geometry cannot be done with high accuracy and our knowledge of the fine-scale organization of deep structures is consequently limited.
The decades-old neurophysiology "gold standard" method for localizing microelectrode recording sites is the electrolytic microlesion—electrical current is used to produce tissue damage at the tip site that can be observed postmortem. Although useful, this method has severe limitations: it damages the tissue of interest, is very labor intensive, and has limited accuracy, especially in 3D (because lesions can be >200 µm, are often irregular in shape, and because tissue must be fixed and cut). Most critically, the method allows only a very limited number of sites within a local region to be reconstructed (
10) and only postmortem. The recent use of microelectrode coating dyes (DiCarlo et al. 1996
; Naselaris et al. 2005; Snodderly et al. 1995) in combination with microdrive depth readings can avoid tissue damage and enhance the number of sites recovered (100), but still incurs the other limitations of microlesions. Some existing imaging modalities can be used on-line and thus can, in principle, allow the reconstruction of an unlimited number of sites. However, these either have limited accuracy (ultrasound; Glimcher et al. 2001) or are costly, labor intensive, and not practical for everyday use (magnetic resonance imaging [MRI] with electrodes in place; Matsui et al. 2007; or following iron deposition; Fung et al. 1998). Extrapolation methods (e.g., frameless stereotaxy), although useful for some surgical procedures, have limited accuracy because small errors are amplified in the process of extrapolation, especially in deep brain structures (as described earlier).
X-rays represent an attractive alternative technology for localizing microelectrodes because electrode materials are typically much more X-ray opaque than biological tissue and because X-ray sources and detectors are commercially available, inexpensive, and robust. Nahm and colleagues (1994) previously described a method based on X-ray stereophotogrammetry to localize microelectrodes during an experiment. In this technique, two X-ray radiographs were taken serially from two vantage points with an electrode in the brain. By locating the microelectrode on each of the resulting X-ray films, the electrode tip position could be triangulated. Additionally, fiducial markers with both X-ray and MRI contrast were affixed to the skull and triangulating these objects allowed coregistration of the X-ray and MR images. However, although the described system was very promising, it had several drawbacks. For one, images were acquired serially, requiring greater operator intervention and increasing the odds that the subject was able to move slightly between acquisitions. The system was also film-based, requiring substantial off-line processing to produce a microelectrode position estimate. In addition, a standard X-ray source with a large focal spot was used, limiting the maximum achievable definition possible in the images. Finally, although the authors demonstrated the system in use, they did not provide a systematic, quantitative analysis of typical accuracy, instead stating that the system accuracy was limited by MRI voxel size, at 625 µm.
Here, we extend this work, building a system based on a pair of microfocal X-ray sources and digital X-ray detectors, with the goal of producing highly accurate, rapid, "real-time" estimates of microelectrode position during an experimental session. Because images can be collected quickly and without interrupting ongoing electrophysiology, we were able to acquire X-ray–based microelectrode tip position estimates of every one of several hundred recording sites over the course of a standard electrophysiological experiment. We argue that this approach opens up new potential for high-resolution physiological 3D mapping of brain activity in awake, behaving animals.
|
|
METHODS |
|---|
|
Two rhesus macaque monkeys were used in this study (although all data, with the exception of skull/dura collision data presented in Fig. 7, were taken from one animal). Aseptic surgery was performed to implant a head post and recording chamber (targeting inferior temporal cortex). Animal training, surgical procedures, eye monitoring, and recording methods were performed using standard techniques for awake, primate visual neurophysiology (e.g., Zoccolan et al. 2005). Because these animals also underwent functional magnetic resonance imaging (fMRI), head-post implant materials were modified to be MR-compatible (Op de Beeck et al. 2008) and eye tracking was done using a video-based system (EyeLink II, SR Research), but these modifications are not required for use of the microelectrode localization method presented here. All procedures were done in accordance with the MIT Committee on Animal Care.
Microelectrode recording
Single microelectrode recordings were performed using standard awake monkey visual physiology techniques (Zoccolan et al. 2005). Glass-coated tungsten microelectrodes (total outer diameter, 310 µm; tungsten shank diameter, 150 µm; length, 130 mm; taper angle, 60°; 0.2–0.3 M
; Alpha Omega) were used to record neuronal activity. On each recording day, a single electrode was advanced through a guide tube (15 mm into tissue) placed in a selected grid location. A hydraulic microdrive (Kopf Instruments) was used to advance the electrode and recordings were made at a series of sites in the temporal lobe (10–30 mm of travel beyond the guide tube), typically with a depth separation of 200–500 µm between each site. The microdrive depth reading was recorded at each recording site and used as a means of validating X-ray position estimates (see RESULTS, Fig. 6). To ensure the accuracy of microdrive readings, the hydraulic fluid was routinely refilled and the drive was monitored for signs of leakage. X-ray imaging (see following text) was performed at each recording site.
For the physiological data reported here, the animal performed a simple visual fixation task during which visual stimuli were rapidly presented (as in Zoccolan 2005). Multiunit activity was collected by placing a hardware threshold at approximately 2 standard deviations above the background voltage fluctuations recorded on the electrode and each crossing of that threshold was counted as a multiunit "event" (see Kreiman et al. 2006 for details). The multiunit threshold was set on each day (i.e., each microelectrode penetration) and was held constant for the entire recording session.
X-ray system
A pair of X-ray images was acquired at each recording site using two microfocal X-ray sources (Apogee 5000, 50 kV, 1 mA, 35-µm focal spot, tungsten target; 0.5-mm aluminum filter; Oxford Instruments), each projecting its beam onto a 1-megapixel CCD X-ray camera (Shad-o-Snap 1024 with Remote RadEye2 EV, Premium Grade, Min-R scintillator, pixel width 
50 x 50 µm; 50 x 50-mm field of view [FOV]; Rad-icon Imaging). The overall system layout is shown in Figs. 1 A and 2. The source–detector pairs were affixed to a custom-built gantry that held the sources and detectors from above (no X-ray image occlusion). This gantry allowed the entire apparatus to be lowered, translated, and rotated around the animal's head and also allowed each source–detector pair to rotate independently around a common center point (and then be fixed in place). The two source–detector pairs were rotated (and then fixed) such that the "spread" angle between them was as close to 90° as possible (an angle of 60° was used here), while ensuring that both pairs had a "clear shot" of the microelectrode and external fiducial markers (see following text) and that the source–detector pairs minimally occluded the visual field of the animal (for the experiments reported here, FOV was approximately ±15° with the system in place). Once rotation spread was determined and fixed, the system was calibrated (see following text) and could then be repositioned each day using four degrees of freedom (df; translation and system rotation) without need for further calibration (see RESULTS). Figure 2 shows how the system was physically arranged in an existing physiology setup.
|
|
The total system cost was approximately $40,000, not including machining/labor costs. The most significant components of this cost are the X-ray sources ($6,000 each) and the X-ray detectors ($13,000 each). Because the system is filmless, its operation costs are negligible. A bill of materials is available at http://www.x-runner.org, along with other information and materials regarding this system.
X-ray safety
Human operator and animal safety were ensured in consultation with MIT's Radiation Safety Office. The X-ray beam cone was approximately matched to the size of the detector plate, so that little of the direct beam escaped the space between the sources and detectors, and lead curtains were placed around the sides of the monkey setup (Fig. 2) behind the detectors as an added precaution against operator exposure from stray X-rays. Our existing monkey setups were additionally already contained with separate "cubicle" rooms (made of drywall, with a standard steel door) and the wallboard and doors in these rooms were already sufficient to reduce exposure to undetectable levels so that no additional room shielding was needed. The X-ray sources and detectors were always operated from outside the cubicle room with the door closed. As a double check, radiation safety badges were placed at fixed locations inside and outside the cubicle room and checked every 3 mo. The measured additional dosing inside the cubicle room averaged only about 0.8 mSv per year during peak usage periods (for comparison, humans naturally receive 1 to 3.6 mSv per year; Bashore et al. 2001; Clarke et al. 1989). To put this in perspective, if a person had been standing in the cubicle room (i.e., a few feet from the animal) for an entire year of active system use (neither needed nor recommended), he or she would have received only slightly more additional dosing than a frequent flyer (0.4 mSv) and less than an airline pilot (Clarke et al. 1989). No radiation dosing above background was found at any tested location outside the room (i.e., no extra radiation at any human locations).
The biggest short-term (nonstochastic) risk to research animals in this system is the potential for skin burning (similar to a sunburn due to absorption of some X-rays at the skin locations facing the sources). For the system configuration used to collect the data reported here, the system delivered about 0.01 Gy to the skin per X-ray image pair acquisition. If 10 such acquisitions are taken per day, the total monthly skin dosing (2 Gy) is less than that which can be used safely in (single-day) human X-ray diagnostic procedures such as fluoroscopy-guided catheterization (Vlietstra et al. 2004). Further optimizations such as thicker source filters (we used a 0.5-mm Al filter) would further harden the X-ray beam (higher fraction of high-energy X rays), which could allow useful images to be acquired with even lower skin risk. Over the course of the experiments presented here (peak usage of 250 acquisitions/mo), no signs of skin erythema or other morbidity were observed in either animal.
External fiducial markers
To provide a common 3D reference frame across sessions, a specially constructed fiducial marker panel (4.5 x 4.0 cm, 6 mm thick) was constructed so that, at the start of each recording session, the panel could be reproducibly placed alongside the animal's head by attachment to the head-post implant (i.e., fixed in a skull-based reference frame). This panel was constructed from plastic (polyetherimide). It was machined such that six, 1-mm-diameter brass spheres (Brass; 1 ± 25.4 µm; Grade 200; Bal-tec) were permanently fixed into holes spaced along the surface of the panel (X-ray–visible fiducial markers) and four 2 x 2-mm cylindrical wells along the surface of the panel were filled with a CuSO4 solution (MR-visible fiducial markers). The precise 3D relationship of these fiducials was needed to accurately project 3D locations into MR volumes (see RESULTS), which was determined using off-site, microcomputed tomography (micro-CT) of the marker panel (Micro Photonics, Allentown, PA) before the experiments began.
Standard workflow
An outline of the standard workflow for our system is shown in Fig. 3. The system is periodically calibrated, using a known standard "calibration object," to precisely estimate the internal system geometry (see following text). When an electrode is positioned in the brain and its 3D tip location is desired, the following steps are performed by the system. 1) Image acquisition: two images (one from each detector) are simultaneously acquired. 2) Image segmentation: corresponding features (fiducial sphere centers and microelectrode tip) are located in each X-ray image. 3) 3D reconstruction: 3D positions of each object (fiducial centers and microelectrode tip) are computed in the native system reference frame using the system's internal geometry determined during calibration. 4) The microelectrode 3D tip position is projected into the common, fiducial-based reference frame (e.g., a skull-based reference frame Fig. 1B). 5) Optional coregistration to other imaging modalities is done using rigid 3D transforms (e.g., to an anatomical MRI image to establish tissue-relative mapping). Images, 3D coordinates, and MRI-coregistered position estimates are displayed to the user for real-time use (e.g., to guide immediate action during a recording session) and stored for off-line analysis.
|
IMAGE PROCESSING AND SEGMENTATION. Gains and offset values were applied to each image pixel based on initial detector sensitivity calibration. Image features (e.g., the microelectrode tip) were then located in each X-ray image pair. For this study, the (x, y) coordinates of the fiducial spheres and the microelectrode tip were located in each image by an experimenter using an in-house, graphical user assist tool; in principle, this step could also be performed using appropriate machine vision techniques (see DISCUSSION). In this work, correspondences between the fiducial projections in each image were selected manually, although this step could, in principle, be done automatically.
3D RECONSTRUCTION IN NATIVE SYSTEM REFERENCE FRAME.
Three-dimensional coordinates of all relevant objects (fiducial sphere centers and electrode tip) were found using standard stereophotogrammetry techniques (Valstar et al. 2002). Briefly, the 3D location of a target was found by computing the best-fit point to the intersection of the rays projecting from each source to the corresponding locations of the feature on each X-ray detector panel. Given the 3D location of the X-ray focal spot (s), the center of the detector panel (d), the unit-length vector normal to the plane of the X-ray detector panel (n), and a known 3D target object location (t; e.g., the center of a small, spherical fiducial), the image of the target will project onto the detector panel at the 3D point P, where
 | (1) |
This point P can be straightforwardly projected into the 2D space of the detector imaging plane, given the 3D location of the upper left corner of the detector panel (u), and the orthonormal column vectors defining the row (r) and column (c) directions of the sensor panel
 | (2) |
Given the preceding equations, one can construct a "forward" projection function F(t, G) that maps the 3D target object position (t) and a collection of parameters describing the system geometry for both source detector pairs (G) onto the vector of image coordinates in both detector panels
 | (3) |
Thus the 3D position of the target point can be estimated by finding the least-squares solution for t in Eq. 3, given both the system geometry G (see following text), and the measurements of P1 and P2 taken from the first and second X-ray images, respectively. In the present work, the Levenberg–Marquardt algorithm was used to perform this optimization and a stable solution was typically found in <10 iterations.
3D RECONSTRUCTION IN FIDUCIAL-MARKER–BASED REFERENCE FRAME. By measuring 3D positions of at least three noncollinear, rigidly arranged fiducial markers, it was straightforward to define a standard reference frame and to project all microelectrode tip position estimates into this frame. For the data collected in this study, the fiducial-marker reference frame was external to the animal and rigidly affixed to the skull (see external fiducial markers in previous text). Thus all microelectrode positions (across recording time within a day or across separate recording days) were reconstructed in a common, skull-based reference frame (the external fiducial reference frame). Internal, tissue-based fiducial reference frames are also possible using the same methods (see DISCUSSION).
COREGISTRATION TO OTHER IMAGING MODALITIES. In principle, microelectrode locations can be determined in any volumetric data that can be coregistered with the X-ray–visible fiducial reference frame (above). In this study, we coregistered to anatomical MRI data. To do this, an anatomical MRI volume (MPRAGE, 448 x 448 x 160, 500-µm isotropic voxels; FOV = 224; Siemens Trio Tim 3T; custom-built surface coil) was acquired of the animal's head with the fiducial panel (described earlier) affixed to the animal's head post (same as during the microelectrode recording sessions). Copper sulfate wells in the fiducial panel produced bright spots on the MRI, establishing an MR-fiducial frame. Given the known (via micro-CT, above) 3D relationship of the (rigid) MR-fiducial reference frame (CuSO4 wells) and the (rigid) X-ray–fiducial reference frame (brass spheres), microelectrode positions in the fiducial reference frame were simply projected into the MR volume using the appropriate rigid-body transformation.
System calibration
System geometry (G in Eq. 3) was precisely determined by imaging a calibration object and system accuracy was measured using a validation object. Both objects were specially fabricated: they consisted of stainless steel spheres (series 440C stainless; 508 ± 2.54 µm, Grade 25; Bal-tec) embedded in a low-temperature coefficient, X-ray–transparent epoxy resin (Epotek 301; Epoxy Technology, Billerica, MA), with the calibration object containing 44 spheres and the validation object containing 6 spheres. The objects were volumetrically imaged off-site using micro-CT (Micro Photonics). The validation object was roughly 6.0 x 5.4 x 7.3 mm and the calibration object was roughly 6.9 x 10.1 x 11.9 mm in size (see Fig. 4), approximating the target imaging volume in the monkey, although this is not required. The known arrangement of fiducial spheres in the calibration object were used in combination with X-ray images of the object to precisely estimate parameters of the internal spatial geometry of the system. In analogy to the solution for 3D target location given known system geometry described earlier (Eq. 3), system geometric parameters (G) were estimated by solving the comparable system of equations for unknown system geometry, given (all 44) known target locations.
|
In all, 12 parameters are needed to fully define the internal system geometry (G): 18 parameters to define two point X-ray sources and two objects (detector arrays) in an external (e.g., room-based) reference frame (minus 6 parameters when the reference frame is specified internally). For clarity, those 12 values are not reported here. However, to give a sense of the system geometry based on such calibration: the distance from the sources to detectors was about 425 mm (pair 1: 399.46 mm; pair 2: 453.51 mm), the spread angle between the two source–detector lines was 59.1° and the approximate magnification of the center of the target imaging area (crossing point of the source–detector lines) was x1.39 (source–detector pair 1) and x1.71 (source–detector pair 2).
System validation
To test system ex vivo performance, stereo X-ray images were taken of the validation object (see preceding text) and the 3D locations of all six fiducial spheres were reconstructed (as described earlier). To obtain accuracy measurements that were comparable to the intended in vivo use, one of the six fiducial spheres was treated as the "microelectrode" and the others were treated as the fiducial reference frame (all combinations were considered in this manner), and the known 3D relative locations of the fiducial spheres (known via micro-CT, above) were used to compute accuracy (see RESULTS).
|
|
RESULTS |
|---|
|
As a first step in determining the accuracy of our 3D X-ray imaging system, we conducted a series of ex vivo tests designed to assess the limits on the system imposed by imaging resolution, X-ray focal spot size, and stability of system geometry (e.g., stability of the spatial arrangement of the X-ray sources, relative to the detectors). A specially created validation object containing six stainless steel spheres (see METHODS) was volumetrically imaged offsite using micro-CT (Micro Photonics) with a resolution of about 5 µm and these micro-CT–derived measurements served as ground truth for the comparisons that follow. The validation object was imaged using our system and the 3D locations of each sphere in the validation object were independently estimated (see METHODS). To compare these estimates with the micro-CT data, a subset of the sphere center position estimates (three, four, or five spheres) was first used to determine an optimal rigid transformation between the micro-CT reference frame and the native (internal geometry) reference frame of our system. The remainder of the sphere center position estimates were then projected into the micro-CT reference frame and errors were computed between the projected estimate and the ground-truth position (error was measured as absolute 3D distance). Figure 5 A shows a summary of the position errors measured by this procedure. In all cases, our system produced highly accurate position estimates, with median error <30 µm and 95% of errors <60 µm.
|
In vivo measurements
To characterize the properties of the X-ray system in practice, we used it to localize 3D microelectrode positions over the course of a standard primate visual electrophysiology experiment. Serial multiunit recordings were made at 228 sites in the temporal lobe, across 20 separate microelectrode penetrations. At each recording site, an X-ray image pair was acquired and the 3D position of the microelectrode was computed (see METHODS). Figure 6 A shows all 228 recording sites registered into a common reference frame (external, skull-based fiducial reference frame; see METHODS). A volume-rendered, coregistered structural MRI image is overlaid in Fig. 6B to provide anatomical bearings.
|
To begin to quantify these factors, we compared X-ray–derived sequential depth estimates to those obtained from the microdrive used to advance the microelectrode. Although microdrive-derived estimates are not especially useful for determining absolute 3D position (e.g., because of the effect of angular errors described in the INTRODUCTION), such measurements are nonetheless quite accurate for the measurement of depth along the microelectrode track. The stepper motor associated with the hydraulic microdrive used here is accurate to 
1 µm and, leaving aside the possibility that the microelectrode curves slightly within the brain and possible errors due to the hydraulics themselves (e.g., leakage, thermal expansion), we here take these microdrive-derived depth measurements to be ground truth.
To compare X-ray–based depth measurements to microdrive-based measurements, a 3D line was fit to the position estimates for the sequence of recording sites along each microelectrode penetration. X-ray–based position estimates were then each decomposed into a depth component (in the direction along the microelectrode track) and a lateral component (absolute deviation from the fit line; Fig. 7A). Since microdrive depth readings are inherently relative (there is no clear "zero" point), the mean was subtracted from the microdrive-derived depth measurements and the mean of the X-ray–derived depth estimates was added, bringing the two sets of estimates into a common frame of reference.
|
Comparison with anatomy
DURA/BONE COLLISIONS. At the end of a few recording sessions, the microelectrode was purposely driven beyond audibly active brain tissue, just into the dura/bone beneath the ventral surface of the temporal lobe. Such dura/bone collisions can provide a rough check on microelectrode depth, since driving a microelectrode into dura mater or bone damages the microelectrode tip, producing a characteristic "crunching" noise when the microelectrode signal is played over a loudspeaker, and an abrupt change in microelectrode impedance. The 3D reconstructed positions of two such collision events are shown in Fig. 8, overlaid on corresponding MR images. The X-ray–determined locations of these collisions on the MR are very near to the ventral edge of the cortical tissue, consistent with the expected collision tissue location.
|
|
|
|
DISCUSSION |
|---|
|
The technique described here substantially extends previous work (Nahm et al. 1994). In contrast to many previously described microelectrode localization techniques (see INTRODUCTION), the method described here is practical for many neurophysiology laboratories, can be deployed on a day-to-day basis, and is minimally intrusive to ongoing experiments. Indeed, the required X-ray images can easily be acquired in a few seconds while a recording session is in progress. Each microelectrode recording site can be individually localized, rather than relying on a single measurement per day, resulting in a fuller picture of each recording session. Also, given that awake electrophysiology is already a highly demanding experimental technique, the relatively small incremental effort required by this technique does not present a substantial additional burden and can be added as a general matter of course. Furthermore, since this technique develops a detailed 3D map of recording sites in vivo and on-line, one does not need to wait for an animal to be euthanized to verify recording sites, and one can "course correct" midstudy to ensure that a given region is properly targeted or sampled with appropriate uniformity.
One immediate application of our technique is in the comparison of electrophysiological data with other measurements of brain activity, such as functional MRI. In particular, given the increasing popularity of fMRI in nonhuman primates, there is an increased demand to reconcile this relatively new technique with traditional direct measures of neuronal activity using microelectrodes. By allowing microelectrode recording locations to be mapped relative to an MR image, spatially resolved single- and multiunit recordings (and local field potentials) can be directly compared with fMRI activity at comparable positions. Such comparisons hold the potential to resolve important outstanding issues in our understanding of how fMRI activity relates to underlying neuronal activity.
More broadly, our technique enables an entirely new kind of maps of neuronal activity to be collected. By collecting large numbers of spatially resolved microelectrode recordings, one can build up highly detailed 3D maps of brain activity with single-unit response specificity. Such maps have the power to reveal details of cortical microstructure (e.g., columns, laminar differences) even in deep brain regions where such structures are currently very difficult or impossible to measure. Knowledge of the fine-scale organization of cells within such brain regions, in turn, has the potential to reveal important insights into their function. It should also be noted that the technique described here need not be limited to recording microelectrodes—stimulating electrodes, electrodes for measuring chemical properties, as well a variety of other invasive devices, such as cannulae (e.g., for drug delivery/measurement, viral vector delivery, etc.) and fiber optics (e.g., for measuring or perturbing neuronal activity with light) could all be localized using this technique.
Future directions
Although the system as instantiated in this study has key features that make it practical and relatively easy to use (see above), and it already has accuracies that compare favorably to those of other techniques, we see several open avenues for improvement.
In terms of accuracy improvement, we first note that, although we here used microdrive readings as an independent measure of accuracy (see RESULTS), in practice, microdrive data can easily be recorded alongside X-ray measurements, and these two data sources can be combined for even greater accuracy. In such a scenario, X-ray data would be used to determine a line fit for the path of the electrode and microdrive depth measurements would then be projected onto this line. Such combined data are particularly useful, since each source of data complements the other. Although microdrives provide no information about the direction in which the microelectrode is tracking, they nonetheless produce highly accurate estimates of distance traveled along that track. Likewise, although the X-ray–derived estimates are somewhat weaker in estimating depths (Fig. 7), they are exquisitely accurate for estimating microelectrode track direction. Although it is not possible from our data to measure directly how much such an integration would improve accuracy, simple simulations of the electrode path line-fitting procedure (assuming 1-µm error in microdrive measurements and X-ray lateral and depth errors of the magnitude demonstrated earlier) suggest that a threefold improvement in overall position accuracy should be possible.
Second, since image quality around the microelectrode tip is an important determinant of overall system accuracy in vivo, any refinement of this technique that improves image contrast should also improve localization accuracy. Perhaps the most obvious candidates for such improvement are the X-ray sources and X-ray detectors. Microfocal sources are currently available with twice the X-ray flux (2-mA current) and a smaller focal spot size (25 µm) compared with the sources used here, which would result in higher image contrast and sharpness, even with shorter imaging time (less chance of motion blur) and more source filtration (for improved animal safety). Given that low image contrast in regions with dense skull interference is a major source of error in the results presented here, we believe that higher flux, in particular, could lead to substantial improvement in accuracy. Likewise, the use of more sensitive X-ray detectors could improve estimates of the microelectrode tip position. Another, related approach would be to enhance the estimate of the microelectrode tip location. For instance, it should be possible to mark microelectrodes at a known distance (e.g., a few millimeters) from their tips (e.g., a notch, bead, or some other feature could be affixed to or scored into the microelectrode shaft). It would then be possible to 3D localize this easier-to-localize feature and then to compute where the tip is relative to that feature (i.e., short distance extrapolation). Along a similar vein, general improvements in software image processing (e.g., filtering) have the potential to improve system accuracy.
Third, tissue distortion is another important factor limiting effective system resolution. Since we are ultimately interested in mapping tissue properties (e.g., neuronal activity), distortion of tissue (e.g., compression as the microelectrode advances) will lead to errors in the tissue-relative 3D maps produced by our technique. Since relatively little quantitative data are available concerning the mechanical behavior of brain tissue during microelectrode recording, it is difficult to know to what extent such compression affects our maps. One way to both study and potentially overcome this limitation would be to implant small fiducial markers "floating" within the brain tissue. Just as chronically implanted flexible microwires are thought to move with brain pulsations (providing more resilient cell isolations; e.g., Porada et al. 2000), small, biocompatible metal fiducial spheres could be embedded in brain tissue such that they provide a marker of brain movement as a microelectrode is advanced. In addition to providing data on the mechanical properties of brain tissue in vivo such fiducial spheres could also potentially serve as an elastic frame of reference that is resilient to tissue distortion (because the frame of reference moves with the tissue). The development of such "internal" reference frames is an area of ongoing research within our laboratory.
Going forward, it will be important to provide even further benchmarks on the accuracy of this system. This is challenging in part because the accuracy we have achieved exceeds the 3D accuracy of readily available methods (e.g., microlesions). At the time of writing, the animals used here are still part of ongoing studies, so we were not able to show a comparison of this technique with histological analysis of microlesions. However, it should be noted that information gleaned from such a comparison is somewhat limited. It is generally not possible to generate accurate 3D reconstructions of brain tissue from histology, due to tissue shrinkage and distortion and due to the nontrivial nature of lining up serial brain slices. One area where microlesion histology does excel, however, is in determining electrode position relative to cortical layer. Assessing the suitability of this technique for estimating the cortical layer of particular recording sites (using either pure X-ray measurements or X-ray plus microdrive) will be an important and interesting future direction.
In terms of further improvements in ease of use, we envision improvements to both the hardware and software that will further enhance the robustness and practicality of this system. On the hardware side, the construction of gantries suitable for holding the sources and detectors in a larger number of configurations is possible and would allow flexible adaptation to the particulars of one's experimental setup and physical constraints. Electronics for full software control of sources and detectors (power, shutter, acquire, safety, etc.) is straightforward and now in place in our laboratory. Fully integrated software for managing calibration, workflow, automated object (fiducial and microelectrode) segmentation, and results visualization is also relatively straightforward. Together these improvements yield a system that is easy-to-use and, whenever the user desires it, produces a high-accuracy, "on-line" electrode 3D localization result in a matter of seconds.
|
|
GRANTS |
|---|
|
|
|
ACKNOWLEDGMENTS |
|---|
|
|
|
FOOTNOTES |
|---|
Address for reprint requests and other correspondence: J. J. DiCarlo, McGovern Institute for Brain Research and Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, MA 02139 (E-mail: dicarlo{at}mit.edu)
|
|
REFERENCES |
|---|
|
Clarke SH, Southwood TRE. Risks from ionizing radiation. Nature 338: 197–198, 1989.[CrossRef][Web of Science][Medline]
DiCarlo JJ, Lane JW, Hsiao SS, Johnson KO. Marking microelectrode penetrations with fluorescent dyes. J Neurosci Methods 64: 75–81, 1996.[CrossRef][Web of Science][Medline]
Fung SH, Burstein D, Born RT. In vivo microelectrode track reconstruction using magnetic resonance imaging. J Neurosci Methods 80: 215–224, 1998.[CrossRef][Web of Science][Medline]
Glimcher PW, Ciaramitaro VM, Platt ML, Bayer HM, Brown MA, Handel A. Application of neurosonography to experimental physiology. J Neurosci Methods 108: 131–144, 2001.[CrossRef][Web of Science][Medline]
Kreiman G, Hung CP, Kraskov A, Quiroga RQ, Poggio T, DiCarlo JJ. Object selectivity of local field potentials and spikes in the macaque inferior temporal cortex. Neuron 49: 433–445, 2006.[CrossRef][Web of Science][Medline]
Matsui T, Koyano KW, Koyama M, Nakahara K, Takeda M, Ohashi Y, Naya Y, Miyashita Y. MRI-based localization of electrophysiological recording sites within the cerebral cortex at single-voxel accuracy. Nat Methods 4: 161–168, 2007.[CrossRef][Web of Science][Medline]
Nahm FK, Dale AM, Albright TD, Amaral DG. In vivo microelectrode localization in the brain of the alert monkey: a combined radiographic and magnetic resonance imaging approach. Exp Brain Res 98: 401–411, 1994.[Web of Science][Medline]
Naselaris T, Merchant H, Amirikian B, Georgopoulos AP. Spatial reconstruction of trajectories of an array of recording microelectrodes. J Neurophysiol 93: 2318–2330, 2005.
Op de Beeck HP, Deutsch JA, Vanduffel W, Kanwisher NG, DiCarlo JJ. A stable topography of selectivity for unfamiliar shape classes in monkey inferior temporal cortex. Cereb Cortex 18: 1676–1694, 2008.
Porada I, Bondar I, Spatz WB, Kruger J. Rabbit and monkey visual cortex: more than a year of recording with up to 64 microelectrodes. J Neurosci Methods 95: 13–28, 2000.[CrossRef][Web of Science][Medline]
Snodderly DM, Gur M. Organization of striate cortex of alert, trained monkeys (Macaca fascicularis): ongoing activity, stimulus selectivity, and widths of receptive field activating regions. J Neurophysiol 74: 2100–2125, 1995.
Valstar ER, Nelissen RGHH, Reiber JHC, Rozing M. The use of Roentgen stereophotogrammetry to study micromotion of orthopaedic implants. Photogram Remote Sens 56: 376–389, 2002.[CrossRef]
Vlietstra RE,Wagner LK, Koenig T, Mettle F. Radiation burns as a severe complication of fluoroscopically guided cardiological interventions. J Interv Cardiol 17: 131–142, 2004.[CrossRef][Medline]
Zoccolan D, Cox DD, DiCarlo JJ. Multiple object response normalization in monkey inferotemporal cortex. J Neurosci 25: 8150–8164, 2005.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Visit Other APS Journals Online |
TOP
ABSTRACT
INTRODUCTION
