Functional Activation Images

In the previous section we described a procedure for analyzing anatomical data in a standardized reference system, often called a stereotaxic system, by morphing all brain images into shape conformation with an anatomical template such as a brain atlas. The same approach can be used for analyzing functional activation images, such as PET or fMRI images. Perhaps the most widespread approach for functional image analysis is based on statistical parametric mapping [37]. In this paradigm, images from many different individuals are first merged together, and regions that are consistently activated during a particular task across subjects are identified through elaborate statistical analysis methods. In order, however, to properly combine functional images of different subjects, we first need to remove anatomic variability. In [37], gross morphological differences across subjects were accounted for using a global polynomial transformation maximizing the similarity of the spatially normalized functional images.

We have taken a different approach, which can potentially improve the registration accuracy of the spatial normalization procedure. In particular, we determine the spatial normalization transformation from the relatively higher resolution anatomic images of a subject. These images are first coregis-tered with the functional images via a rigid body registration method such as the ones in [38,39]. The elastic warping of the anatomic images is then determined, as described in the previous sections. The same transformation is finally applied to the coregistered functional images, mapping them to the stereotaxic space.

Figure 12 demonstrates the procedure for mapping a PET image to the Talairach stereotaxic space. The spatial transfor-

FIGURE 13 (Left, right) Two spine images of two different individuals. (Middle) An elastic transformation of the image on the left that brings the spine into registration with the one on the right. If a number of images are spatially transformed to match the same template, then correlations between the location or size of a spinal lesion and concomitant clinical symptoms, such as pain, can be readily calculated.

FIGURE 13 (Left, right) Two spine images of two different individuals. (Middle) An elastic transformation of the image on the left that brings the spine into registration with the one on the right. If a number of images are spatially transformed to match the same template, then correlations between the location or size of a spinal lesion and concomitant clinical symptoms, such as pain, can be readily calculated.

mation here was determined from the corresponding anatomic image, whose axial resolution was 1.5 mm. Overlaid on the spatially normalized PET image is shown the atlas associated with the Talairach space, showing a good spatial correspondence between the morphed PET image and the target template image. Our experiments have shown that a substantial increase in the accuracy and sensitivity of activation focus detection can be achieved this way [40].

0 0

Post a comment