click here for PDF of NeNa-poster at Human Brain Mapping 2003

Introduction

TMS is a new, safe, non-invasive and reversible method for intervening with neural processes in the human brain by means of a short magnetic pulse, in order to study the involvement of brain areas in certain types of behavior or actions. A TMS device consists of a coil through which a condensator can discharge a strong current, generating a short-lived magnetic induction field perpendicular to the coil's surface. When the coil is held over the surface of a participants scalp, the magnetic pulse will hus be directed into the the head, directed at the cortex (see figure 1 for a possible TMS setup). TMS delivers short (<1 ms) magnetic pulses up to 2.5 T that penetrate the skull and dirsupt neural processing for a very brief period (Walsh and Cowey 1998). At present, two methods for TMS administration are in use. The repetetive TMS (rTMS) mehod offers a train of pulses for an extended period of time (several minutes). A relatively large brain area will be affected up to half an hour, creating a 'virtual lesion'. Testing the participant can reveal functional deficits similar to neuropsychological studies. In single pulse studies, only 1 pulse is administred, which will induce changes in neurophysiology for tens of milliseconds, after which the neural system again functions normally. The affected brain area is considerably smaller than in rTMS studies. The latter more subtle single pulse method also allows one to investigate the time course in which a brain area is contributing to cognitive processes. 

For single pulse TMS, the placement of the coil above the proper site of the cortex is quite critical. In the past, researchers often placed the TMS coil relative to the positions of EEG cap electrode positions, yielding locations as known from average brain anatomy. Also, it is common practice to first identify the location on the scalp over the primary motor cortex where TMS can evoke small thumb movements (Paus et al, 1999), the so called 'motor hotspot' (see figure 2). Relative to this position, the coil is then shifted to a region of interest, whose position is again obtained from a brain atlas, based on average brain coordinates. Several serious problems arise when using both aforementioned methods, in that one doesn't know exactly which brain area is affected by the TMS pulses. The anatomy of the central nervous system differs substantially between individuals, so the method described above only gives a rough estimation of the spot being searched for. As a result, one might not stimulate the actual area of interest. Second, even with a similar brain architecture, functional activations of neural tissue may be located at different locations due to brain plasticity, an effect that is known to exist in the human brain. Therefore, even when the correct area is stimulated, it is not guaranteed that the intended functionality is affected. An accurate search for the desired stimulus location therefore requires both knowledge about the exact anatomy of the test subjects brain and data about the functional activity of his brain.

Therefore, we constructed NeNa (Neural Navigator), a frameless stereotactic device which allows us to use a structural MRI scan of a persons head to guide the the TMS coil to the proper region on the skull. On top of that, it is possible to superimpose functional MRI (fMRI) activation maps on the structural brain scan, to allow researchers to target functional areas in exactly the same individuals that participated in an earlier fMRI study. By doing so, the probability of affecting the intended neuroanatomical structure, and even the intended functions, is much greater.

The main problem is to align the 3D space (of a 3D position measurement device) the participant's head is in to the MRI coordinates obtained earlier. In other words, one wants to know where the TMS coil, that the experimentator is holding near the scalp of the participant, is located with respect to the brain structure of the participant on the computer screen. In order to be able to guide the placement of the coil by the MRI image, one has to transform the 3D coordinates of the coil, as measured by an 3D position measurement device attached to it, to MRI voxel coordinates. When the transformation is calculated, one can then draw a 3D representation of the coil together with a 3D representation of the brain surface on the computer screen in real time, i.e. one can see a 3D drawing of the coil 'move' with respect to the brain on the screen.

In order to do be able to calculate the position of the coil in 3D MRI voxel coordinates, we adopted a method that uses anatomical landmarks that are visible in both the MRI scan and on the surface of the participants head, such as the bridge of the nose, the ears, etc. There is no need to place markers on the participants head during the scans, although it could of course increase precision. First, in the NeNa rendering window 3D cursors are moved to these anatomical landmarks on a skin rendering of the particpants MRI scan, and thier MRI coordinates stored on disk. With the 3d position measurement device MiniBIRD these 3D positions will are measured at the head of the participant and stored on disk, and finally a 3D affince transformation will be calculated that can map the measured BIRD marker positions to the MRI scan's marker coordinates as accurately as possible. The latter is not straightforward, since all measurements are subject to small errors, that shouldn't alter the calculated mapping too much. In other words, the transformation algorithms should be robust with respect to small deviations in the measured coordinates of anatomical landmarks. Below, some more information on the used algorithms, implementation and visualization of the MRI data, as well as some first results (NeNa screenshots) are presented.

Transformation algorithms

The main problem to be solved is depicted in figure 3, in 2 dimensions. A set of points B_i, representing anatomical landmarks on the surface of the head as measured by the miniBIRD system, has to be mapped onto a set of points M_i , representing the same positions selected in MRI coordinates in the 3D browser of NeNa, as accurately as possible.

The 7 parameters that will have to be estimated are determined in 2 stages. First, a 'smart guess' of each of the translation, rotation and scale factors is determined directly from the stored coordinates B_i and M_i . We developed algorithms to do so, mainly based on the Jacobian transformation matrix. Details will be published in the near future (Neggers & Langerak, 2003, in preparation). The 'smart guess' already yields acceptable mappings. Second, the sumsquared error, SSE, of the difference between B'_i (the transformed  B_i) and M_i will be minimized further by means of a gradient descent minimilazation algorithm (Powell, 1964), with the 7 translation, rotation and scale factors as free parameters. The latter procedure has already proven to even further optimize the mapping transformation. See figure 5 for a screenshot from NeNa with a visualization of B'_i and M_i , lying very close together.

Visualisation

For the visualization of 3D data in NeNa, we use the open source Visualizion ToolKit (VTK), that comes with a complete 3D rendering pipeline and a great many 3D visualization and manipulation functions and algorithms. One of the most important algorithms for the functioning of NeNa is the Marching Cubes algorithm (Lorensen et al, 1987). With the Marching Cubes algorithm and the right paramaters (mainly intensity trhesholds), one can extract polygonal surface data (for example triangles) from volumetric data (voxel data), such as MRI scans. Polygonal data is much easier to visualize, and it takes much less computational effort to render it on the screen. The latter is allowing the change of view on the data and other rendering ascpects in real time, which comes in handy for a tool like a neuronavigator. For more details on the visualization method used in NeNa visit: http://public.kitware.com/VTK/.

Implementation and compatible data types.

NeNa is written in C++ for MS Windows platforms (98, ME, 2000 and XP), using the Microsoft Visual Studio compiler. Additional libraries are used, such as the aforementioned VTK library, and the MiniBIRD library and dll provided by EST Technologies for controlling the MiniBIRD device.

NeNa can read MRI volumetric image data in the Analyze format, developed by the Mayo Clinic. The Analyze format has been adopted as the standard data format by the open source (f)MRI analysis software, SPM, which became an informal standard in the functional neuroimaging community. Therefore, NeNa is completely compatible with the functional maps (Statistical Parametrical Maps) that form the output of an SPM analysis, and can render them on the screen in order to target active regions with a TMS pulse. For a TMS experiment using all the features of NeNa (skin rendering, cortex rendering, and activation maps, see figure 5), one needs to prepare 3 (f)MRI data files in the Analyze format:

1. A structural (T1) MRI image of, if possible, the entire head (at least from the vertex down to the upper-lips) in order to set sufficient recognizable anatomical landmarks. From this volume the skin rendering will be created.
2. A segmented brain image of the so called gray matter, in order to create a rendering of the surface of the brain. Segmenting a strucural MRI data volume means separating the gray matter, white matter and remaining tissue in 3 different volumes. An adequate tool for segmenting a MRI structural data set is present in the recent releases of SPM.
3. A statistical map of brain activity during a certain task, such as a T-map, most likely obtained from an analysis of T2* time-series. This volume will be rendered as blobs in a different color in NeNa, with an adjustable statistical threshold.

Here it is important to stress that these volumes are in so called 'native space', and not normalized to standard average brain coordinates such as Talaraich or MNI. One wants to target activity during a certain task in one specific participant, that was in the scanner shortly before the TMS experiment. A brain structure in normalized brain data of a person is likely to be in very different places as the same structure in native non-normalized brain data, which is at the correct (corresponding to the participants neuroanatomy) location.

Results

For one individual (the first author) the 3 imaging volumes mentioned above are created. The functional map is a T-map testing the difference in regional blood flow between right thumb tapping against rest. When loaded in NeNa, on obtains an image in the rendering window as in figure 5. One can clearly see a surface rendering of the skin (gray opaque), the brain (blue) and activity maps (red). This map is rotatable with the mouse.

Figure 5. Rendered surfaces of the skin (gray opaque), the brain (blue) and activity patterns from a thumb tapping task in the NeNa rendering window. Blue spheres represent the MRI markers as set manually in the rendering window, and yellow markers represent the MiniBIRD marker positions after the discussed mapping transformations from MiniBIRD space to MRI space were applied.

After postioning the 3D sphere-shaped colored cursors in the MRI data in the rendering window (blue spheres), the same 3D positions were measured with the MiniBIRD probe at the participant that is sitting in the TMS setup. Next, the mapping transformations described above were calculated, and applied in NeNa. In Nena, one can draw the transformed MiniBIRD marker coordinates B'_i (yellow spheres), that are indeed very close to the markers set in MRI space in the example in figure 5. When one is not satisfied with the accuracy of the mapping, it can be redone from scratch, or single MiniBIRD markers that have relatively large deviations can be remeasured after clicking on them with the mouse. After a satisfying mapping, one can now move the MiniBIRD probe over the head of the participant, and see where it is located and how it is oriented. The MiniBIRD probe is a device holding the MiniBIRDs receiver unit at the base of cylinder, that is extended into a 2 mm wide rounded tip. Since the miniBIRD receiver at the base also measures, besides position, the orientation, one can calculate where the tip is located. See the big (tip of the MiniBIRD device) and the small red sphere (base of the MiniBIRD device) hovering over the head in figure 6. A thin red line connects both spheres, and extends into the head. When the skin is made opaque as in figure 5, one can see what brain structures and/or activities are targered by the MiniBIRD device, and set a mark on the participants scalp for TMS administration.

Figure 6. The MiniBIRD device hovering over the participants head in NeNa's rendering window, after the mapping transformation has been aplied. One can see the orientation of the device and how it is oriented and hence where an imaginary line  is penetrating the head (representing the TMS pulse).

Finally, NeNa can also be extended to show a sliced view of the MRI scan, besides the surface renderings shown in figure 5. The 3 extra windows in figure 7 show the three orthogonal views on the structural (T1) MRI scan that was loaded, intersecting at the MiniBIRD probe position, i.e. also moving in real time when the experimentator moves the MiniBIRD probe around the head of the surface. Some structures are better identified on slices than on renderings, and another advantage is that clinicians are usually trained to identify brain regions from a sliced view.

Figure 7. The NeNa rendering window can be extended with a more conventional sliced view of the structural MRI data, on which some structures are easier to identify than on a surface rendering.

References

Powell, M. (1964) An efficient method for finding the minimum of a function of several variables without calculating derivatives. Computer Journal, 7:155-162

Paus, T. (1999) Imaging the brain before, during, and after transcranial magnetic stimulation. Neuropsychologia 37:219-224.

Cowey A, Walsh V (1998) Magnetic stimulation studies of visual cognition. Trends in cognitive sciences 2:103-110.

Lorensen, W.E. , Cline, H.E. (1987). Marching Cubes: a high resolution 3D surface reconstruction algorithm. Computer Graphics 21(4):163-169 (Proc. of SIGGRAPH), 1987.

S.F.W. Neggers, R. Langerak, R. Mandl, D. Schutter and A. Postma (2003). (f)MRI guided TMS using frameless stereotaxy: validation of NeNa with a cortical map of EMG responses. Neuroimage 19, 67–88. (poster at HBM 2003, download PDF)