Manual SN Example

The simplest means to transform a brain image to Talairach space is to manually apply rotations, translations, and scaling while visually matching source and target brain features. The 1988 Talairach atlas with its key features for location, orientation, and dimension helps to illustrate this process. The order of rotation and scaling are extremely important since the Talairach brain dimensions only apply to a brain in the standard orientation; therefore, rotations must be done before scaling. The suggested steps for manual SN are as follows:

• Mid-sagittal alignment. The brain is divided into right and left hemispheres by the interhemispheric fissure.

FIGURE 3 (a) Axial views of the brain before (left) and after (right) alignment to the mid-sagittal plane (red line). The image was rotated clockwise and translated to the right in this example. Note the embedded marker used to positively identify the right side of the patient. (b) Coronal views of brain images before (left) and after (right) alignment to the mid-sagittal plane (red line). The brain was rotated counterclockwise and translated to the right in this example. (c) Sagittal views of the brain before (left) and after (right) mid-sagittal alignment. Following alignment (right) sulcal detail is diminished and lower in contrast, indicating better match of this mid-sagittal view to the interhemispheric fissure. See also Plate 78.

FIGURE 3 (a) Axial views of the brain before (left) and after (right) alignment to the mid-sagittal plane (red line). The image was rotated clockwise and translated to the right in this example. Note the embedded marker used to positively identify the right side of the patient. (b) Coronal views of brain images before (left) and after (right) alignment to the mid-sagittal plane (red line). The brain was rotated counterclockwise and translated to the right in this example. (c) Sagittal views of the brain before (left) and after (right) mid-sagittal alignment. Following alignment (right) sulcal detail is diminished and lower in contrast, indicating better match of this mid-sagittal view to the interhemispheric fissure. See also Plate 78.

A n It rio r-Super ior

FIGURE 3 (Continued).

Since this fissure approximates to a plane, the mid-sagittal plane, it serves as an excellent landmark for alignment. The objective of mid-sagittal alignment is to rotate and translate the brain to align the interhemi-spheric fissure of the brain with the y-z or mid-sagittal

plane ofTalairach space (Fig. 3). By convention, the mid-sagittal plane is located in the middle of the left-to-right (L-R) field of view (FOV) of the brain image. Most brain images are acquired with a 256-mm L-R FOV, so the mid-sagittal plane is set at 128 mm from the left edge.

FIGURE 4 Mid-sagittal section views of the brain before (left) and after (right) AC-PC alignment using a four-point fitting method. The four landmarks are the anterior-inferior margin of the corpus callosum (blue), inferior margin of the thalamus nucleus (yellow), the superior colliculus (green), and the apex of the cerebellum. See also Plate 79.

Axial, coronal, and sagittal section views of the brain are used to interactively assess feature matching while adjusting rotation and translation parameters. Graphical lines representing the intersection of the mid-sagittal plane in axial and coronal section images are used as guides (Figs 3a, b). The brain image is translated and rotated until the interhemispheric fissure aligns with the mid-sagittal plane in both images. Avisual indication of mid-sagittal alignment is provided in the sagittal view that displays the interhemispheric fissure when the alignment is correct. In MR images, a broad region with loss of sulcal detail is characteristic of the interhemispheric fissure, and visual assessment of this effect is used as a qualitative measure of alignment (Fig. 3c). Most users have a high level of confidence in mid-sagittal alignment, even for lower resolution PET images. AC-PC alignment. Following mid-sagittal alignment only rotation about the x-axis is needed for standard Talairach orientation. Unlike the large distinct inter-hemispheric fissure, landmarks for this alignment, the AC and PC, are small and the PC is sometimes hard to see, even in high-resolution MR images. Formally, the AC-PC line is the line passing through the superior edge of the AC and the inferior edge of the PC [49]. When the PC is well visualized, the operator can manually identify the AC and PC to fit the AC-PC line. Several investigators [2, 23, 32, 38] have used least-square error fitting methods with additional landmarks characteristic of the AC-PC line. The use of more landmarks to fit the AC-PC line using least-square error methods is less susceptible to operator bias and landmark anatomical variability and helpful in low-resolution images where the AC and/or PC cannot be seen. Figure 4 illustrates the use of a four-landmark method [32]. Once the AC-PC line is determined, its angle with the y-axis is calculated and the image rotated and translated so that the AC-PC line is congruent with the y-axis.

Dimension normalization. Bounding box dimensions of the brain are measured following standardization of brain orientation by mid-sagittal and AC-PC alignment. The maximum extent of the brain in x, y, and z directions is used for its bounding box dimensions (Fig. 5). These can be determined visually in high-resolution MR images or in low-resolution images such as PET, with the aid of boundary detection methods [32]. The visual determination of brain extent in MR images is best done while viewing sections perpendicular to the direction of interest. The location of the bounding limit is the coordinate of first section image where the brain is visualized. This is relatively simple for the front (+y), back (—y), left (—x), right (+x), and top (+z) brain extent. However, the inferior brain extent ( — z), identified as the inferior margin of the temporal lobe, is surrounded by many confounding structures (Fig. 5b).

By careful inspection of axial, coronal, and sagittal views the inferior boundary can be accurately identified. Once the bounding box dimensions are measured they are compared with those of the 1988 Talairach atlas brain to determine scale factors. Scaling is then applied independently along each axis such that the transformed brain's bounding box dimensions are identical to those of the atlas brain.

• Origin placement. This step can be done before or after dimension normalization, but the end result must be a precisely designated location within the image volume for the origin or AC. This is readily done visually and the image translated such that the AC falls at a designated location (see Fig. 6). In low-resolution images, where the AC is not visible, a secondary method for origin placement assumes that the AC is 40% from the anterior to the posterior boundary of the brain. This approximate site for the AC was the result of measurements in numerous brain images following global Talairach spatial normalization [32].

For global SN methods, the accuracy depends on location, size, and anatomical variability of the site of interest. For example, the anterior commissure, a small reference landmark in many global SN methods, will match very well (Fig. 6). Likewise, larger anatomical structures near the AC and PC tends to match reasonably well (Fig. 7). However, match quality generally diminishes with distance from the center of the brain. Features in the cortex are harder to match because landmarks are difficult to accurately define, and anatomical variability is often greater there. Feature matching for global SN methods has been estimated to exceed 90% by several different methods with that of regional methods reaching 95% [5,6,31,34,57].

The feature matching capability of global SN is less problematic in low-resolution PET brain function studies than in high-resolution MRI anatomical studies. The spatial resolution of imaging systems is determined by the extent of a theoretical point source and quantified by the full-width-half-maximum (FWHM) of the point-spread function (PSF) [12]. Point sources separated by a distance less than the FWHM tend to merge into a single source [30], therefore, a smaller FWHM indicates higher spatial resolution. It is common practice to spatially smooth PET images to reduce noise levels [19], further increasing the overall FWHM and reducing spatial resolution. The net FWHM of PET studies can be greater than 1cm [24], and since global SN usually provides anatomical matching of 1 cm or less in most areas, corresponding anatomical regions tend to merge in group average images

0 0

Post a comment