Neuroimaging, VII: PET and the averaging of brain images

Abstract

P ositron emission tomography (PET) studies provide functional measures of brain activity. Accurate anatomical localization can be provided by aligning a magnetic resonance imaging (MM) scan with the PET scan. Statistical analyses can be carried out on each brain coordinate to provide exploratory maps. Combining different individuals’ brains is analogous to averaging together photographs of different people’s faces; every one has a different shape. To achieve interpretable differences in pupil color, lip size, or nose length, the eyes, mouth, and nose must be placed at the same x and y locations in every image-a process that requires stretching every face to fit some common mold. The top row of images above shows MRI scans of 70 normal subjects averaged together in three different ways. On the left, brain images are averaged on the center of the slice, a process analogous to averaging face photographs by piling them up on the tip of the nose. This spindle method produces a fuzzy image. In the middle column, images are averaged by finding the edges of each slice and stretching it horizontally and vertically so that each subject has the average width and average length. This bounding box method produces a sharper image, especially where the cortical rim touches the box. It is similar to stretching each face so that the chin-to-top-of-head and ear-to-ear distances are the same. The right column uses 500 points in the cortical perimeter and nine midline landmarks to stretch the brain to the average. This linear perimeter-anchor morphing method produces a slightly sharper image. The averaged PET scans in row 2 show the same progression of sharpness. The cortical surface shows the transition from the high metabolic rate of cortical neurons to the low skull metabolic rate as a fuzzy blue-green band (left) because of individual differences in brain size. This imperfection is improved with the bounding box (center) and sharpened with linear perimeter-anchor morphing (right). The bottom row demonstrates progressive improvement most clearly. The color scale depicts large (red) and small (blue) standard deviations. Cortical rim variability is especially prominent for the spindle method (left). Variability is reduced in the bounding box method (middle) but is still noticeable, especially at 2, 5, 7, and 1 1 o’clock, where differences in head roundness are not controlled. The all-blue cortical rim (right) shows the reduction in variability with edge morphing. Statistical tests on the fuzzy image (left) would be relatively insensitive, especially for the cortical surface, where a given location might be deep in the cortex for one person and outside the brain for another. A sharper cortex is seen where edges are morphed to the same contour (right), providing opportunities for more sensitive detection of task activation or patient group differences.