1Christine N. Vidal, 3Timothy J. DeVito, 1Kiralee M. Hayashi, 3Dick J. Drost PhD, 2,3Peter C. Williamson MD, 2Beth Craven-Thuss, 1David Herman, 1Yihong Sui, 1Arthur W. Toga PhD, 2,3Rob Nicolson MD, 1Paul M. Thompson PhD
1Laboratory of Neuro Imaging, Dept. Neurology, UCLA School of Medicine, Los Angeles, CA, USA
2Dept. Psychiatry, University of Western Ontario, London, Canada
3Dept. of Medical Biophysics, University of Western Ontario, London, Canada
We developed a novel computational strategy to detect and map, for the first time, the spatial pattern of corpus callosum abnormalities in autistic children. By contrast with volumetric approaches, we created a statistical brain atlas, based on anatomical surface meshes, to encode morphological variability in the shape and thickness of the corpus callosum in normal and autistic children. Statistical criteria were developed to pinpoint local regions of abnormal callosal thinning. This revealed the spatial pattern of deficits in autistic children, with substantially greater discriminatory power than volumetric measures. The resulting maps may reconcile previously conflicting volumetric findings in autism, revealing the spatial pattern of abnormalities with greater visual and statistical power.
35 T1-weighted 3D MP-RAGE MRI volumes (1.2-mm isotropic resolution) were acquired at 3.0-Tesla (IMRIS, Winnipeg, Canada) from 15 autistic children (mean age: 9.9 yrs.+/-3.2SD), 13 healthy controls (10.0 yrs.+/-2.1SD), and a third group of 7 matched subjects (10.5 yrs.+/-2.8SD) with Tourette syndrome. All subjects were matched for age, height, sex, and other demographic criteria. Scans were normalized by affine transformation to ICBM standardized stereotaxic space (Mazziotta et al., 2001). One rater, blind to age, gender and diagnosis, repeatedly traced the corpus callosum (allowing the quantification of reliability) as follows: both (1) as a single curve in the midsagittal plane (3 times); and (2) as a 3D anatomical surface extending 4 mm either side of the interhemispheric fissure (twice). All traces were uniformly redigitized to render the sampled points spatially uniform, and converted into 3D parametric surface meshes. Surface meshes were averaged across subjects to create average shape models for each diagnostic category, and these were overlaid to reveal systematic group differences in anatomical shape (Fig. 1(c)). Intersubject variations in shape and thickness were encoded using a spatially parameterized covariance field. Thickness maps were plotted on the callosal boundaries by first deriving a medial surface equidistant between the upper and lower surfaces for each subject, and mapping the 3D distance of this medial surface to each boundary point. Thickness maps were averaged across subjects, and group differences were assessed at each surface location by multivariate analysis of variance. Regions exhibiting significant differences were coded in color on the surface meshes (Fig. 1(b)). Mean group differences in callosal anatomy were plotted as a percentage reduction in local thickness (Fig. 1(d)). Permutation testing of the shape statistics was used to assess their significance, while adjusting for multiple comparisons and modeling the autocorrelation of the residuals of the statistical model. The validity and statistical power of the deficit maps was assessed by comparing them to more traditional volumetric measures. The Witelson partitioning scheme (Fig. 1(a)) was used to divide the callosum into discrete sectors, representing the: (1) splenium; (2) isthmus; (3) posterior midbody; (4) anterior midbody; and (5) anterior third (Witelson, 1989). 3D maps were also created to emphasize callosal shape differences between patients and controls, visualizing the regions with greatest effects.
Average maps of callosal anatomy revealed profound reductions in callosal thickness in autism, with most significant effects in the genu, midbody and splenium (Fig. 1(a); corrected significance: p<0.0002, permutation test). After adjusting for age effects, volumetric measures also revealed a 13.7% smaller midsagittal callosal area, overall, in autistic children (604.8+/-37.2 mm2), relative to controls (700.7+/-21.8 mm2; p<0.021). The most significant reductions were identified in the genu (14.8%; p<0.016), midbody (13.3%; p<0.047), and splenium (14.9%; p<0.036). In this sample, no differences in overall brain volume were detected (p>0.38). The results of the Witelson partitioning tests are shown in Fig. 1(a). These area reductions are highly consistent with the significance map of thickness differences between patients and controls. Each algorithm highlights deficits in the splenium, posterior midbody and anterior callosal regions. No shape or thickness differences were found in the medication-matched sample, although expansion of the sample is necessary to further understand the specificity of these effects.
A new anatomical mapping strategy was developed to detect and visualize patterns of callosal thinning in autistic children. Clear deficits were observed, using a parametric mesh modeling technique and a normative atlas encoding statistical variations in surface geometry, shape and thickness. Greatest deficits were identified in the genu, splenium, and midbody. These anomalies may suggest aberrant connectivity or impaired axonal transfer of cognitive and perceptual information between brain hemispheres, in frontal and occipital regions. The fluctuating effect size, and lower power, of volumetric measures may explain why some prior studies have found anatomical reductions in autistic children to be significant only in the genu and rostrum (Hardan et al., 2000), or most prominent in the body or posterior sectors (Egaas et al., 1995; Piven et al. 1997; Manes et al., 1999). Shape mapping provides increased power for group discrimination (p<0.0002), relative to volumetric measures (p<0.02). This increased power, as well as the ability to visualize deficit profiles in the form of a map, may be advantageous for genetic, behavioral, and therapeutic studies of autism. It may also help provide an MRI-based biomarker for autism and related neuropsychiatric disorders.
References: Mazziotta J et al. (2001). J. Roy. Soc. 356 (1412):1293-1322; Witelson S (1989). Brain 112:799-835; Hardan A (2000). Neurology 55:1033-1036; Egaas B et al. (1995). Arch. Neurol. 52:794-801; Piven J et al. (1997). Am. J. Psychiatry 154:1051-1056; Manes F et al. (1999). J Neuropsych. Clin. Neurosci.11(4):470-4.
Paul Thompson, Ph.D.
|RESUME| E-MAIL ME| PERSONAL HOMEPAGE| PROJECTS|