Predictive models of autism spectrum disorder based on brain regional cortical thickness
Introduction
Autism spectrum disorder (ASD) is a neurodevelopmental disorder with a prevalence of approximately in 1 in 150 children (Amaral et al., 2008). Children with ASD have abnormal social behavior, impaired communication and language skills, and repetitive/stereotyped behavior (Belmonte et al., 2004, Nordahl et al., 2007, Rapin, 1997). In studies of cognition, ASD has been defined as fundamental deficits in central coherence, executive function, and empathizing (Belmonte et al., 2004). MR examination has revealed that children with ASD have subtle structural changes in many brain structures, including the frontal lobe, parietal lobe, hippocampus, amygdala, cerebellum and brain stem (Courchesne et al., 2001, Geschwind, 2009, Hashimoto et al., 1995, Muller, 2007, Sparks et al., 2002).
Many morphological studies of ASD have used voxel-based morphometry (VBM), which measures voxel-wise gray- and white-matter volume changes across the entire brain. These VBM-based studies identified both gray- and white-matter volumetric changes (Aylward et al., 2002, Aylward et al., 1999, Ke et al., 2008, Rojas et al., 2006). However, the intrinsic topology of the cerebral cortex is that of a 2-D sheet with a highly folded and curved geometry (Fischl et al., 1999), and VBM cannot directly measure this topology. Surface-based morphometry (SBM), which centers on the computation of cortical topographic measurements, has the potential to provide information complementary to that provided by VBM. SBM can derive features such as regional gray-matter thickness and regional surface area (Voets et al., 2008), as well as curvature and sulcal depth (Fischl et al., 1999, Kim et al., 2005).
There have been several SBM studies of ASD. For example, Hadjikhani et al. (2007) reported thickness differences in the mirror-neuron system and other areas involved in social cognition in individuals with ASD. Nordahl et al. (2007) demonstrated cortical-folding abnormalities in individuals with ASD, primarily in the left operculum, bilateral parietal operculum, and bilateral intraparietal sulcus.
Currently, ASD is diagnosed based on behavioral criteria. Given the VBM and SBM findings described above, an MR-based diagnostic model holds the promise of enhancing, perhaps complementing, behavioral assessment. Toward this end, Akshoomoff et al. (2004) entered six pre-selected brain volume-based features into discriminant analysis and correctly classified 95% of very young people with ASD; however this accuracy rate was based on reclassification of the training set, rather than on cross-validation or classification of an independent test set. Ecker et al. (2009) investigated the predictive value of whole-brain structural volumetric changes in ASD, using SVM classifiers, and obtained 81% classification accuracy based on cross-validation. Singh et al. (2008) developed a diagnostic model generated by the LPboost based algorithm to distinguish autistic children from control subjects, based on voxel-wise cortical thickness, based on approximately 40,000 points for each subject; they reported 89% classification accuracy based on cross-validation. The principal limitation of their work was basing the feature dimension reduction step on all samples outside cross-validation.
In this study, we test the hypothesis that diagnostic models can distinguish children with ASD from controls based on regional cortical thickness, and that these models have greater accuracies than diagnostic models based on regional volumes. To test this hypothesis, we first computed average cortical thicknesses and volumes of 66 structures defined on a brain atlas, for each subject. We then applied four data-mining approaches to generate four diagnostic models based on either regional cortical thicknesses or regional volumes. Finally, we compared performance metrics of thickness-based diagnostic models with those of volume-based diagnostic models.
Section snippets
Participants
Participants in this study, aged 6–15 years, consisted of two groups: 22 children with ASD (mean age, 9.2 ± 2.1 years), and 16 volunteer control subjects (VC) (mean age, 10.0 ± 1.9). Children with ASD and control subjects were group-matched on age, sex, full-scale IQ, handedness, weight, height, and socioeconomic status. All participants with ASD were recruited by the Child Mental Health Research Center of Nanjing Brain Hospital. The diagnosis of ASD was based on the criteria of the fourth edition
Results
Table 1 lists demographic characteristics of subjects in the ASD and VC groups. We did not find significant differences between groups in gender (chi-square p-value = 0.67), age (t-test p-value = 0.25), WISC full-scale IQ (t-test p-value = 0.44), weight (t-test p-value = 0.12), or height (t-test p-value = 0.16).
Table 2A and 2B list classification performance metrics of the four diagnostic models for cortical regional thicknesses, and regional volumes, respectively. Figs. 3A and B show the ROC curves of
Discussion
To our knowledge, there has been no previous attempt to compare thickness- and volume-based classification of ASD. We found that thickness-based classification was more accurate than volume-based classification, for each combination of classifier and performance metric.
We used four machine-learning methods to generate classifiers, in order to avoid bias with respect to the functional form of the classifier. The results listed in Table 2 suggest that (1) SVM and MLP have similar performance, and
Conclusion
To our knowledge, this study represents the first attempt to classify autism using regional cortical thickness measurements extracted from SBM, and to compare classification results to those obtained from volume-based analysis. The principal contribution of our work is our determination that thickness-based classification results in significantly more accurate classification than volume-based classification, across four different classification approaches, and four classification accuracy
Acknowledgments
Yun Jiao was supported by China Scholarship Council 2008101370 and National Natural Science foundation of China, Project No. 30570655. Drs. Chen and Herskovits are supported by National Institutes of Health grant R01 AG13743, which is funded by the National Institute of Aging, the National Institute of Mental Health, and the National Cancer Institute. They are also supported by NIH R03 EB009310. Drs. Ke and Chu were supported by the Natural Science Foundation of Jiangsu, China (No: BK2008082).
References (57)
- et al.
Outcome classification of preschool children with autism spectrum disorders using MRI brain measures
J. Am. Acad. Child Adolesc. Psych.
(2004) - et al.
Neuroanatomy of autism
Trends Neurosci.
(2008) - et al.
Cortical surface-based analysis. I. Segmentation and surface reconstruction
NeuroImage
(1999) - et al.
An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest
NeuroImage
(2006) - et al.
Cortical surface-based analysis. II. Inflation, flattening, and a surface-based coordinate system
NeuroImage
(1999) - et al.
Artificial neural networks (the multilayer perceptron)—a review of applications in the atmospheric sciences
Atmos. Environ.
(1998) - et al.
Reliability of MRI-derived measurements of human cerebral cortical thickness: the effects of field strength, scanner upgrade and manufacturer
NeuroImage
(2006) - et al.
The contribution of neuropathologic studies to the understanding of autism
Neurol. Clin.
(1993) - et al.
Automated 3-D extraction and evaluation of the inner and outer cortical surfaces using a Laplacian map and partial volume effect classification
NeuroImage
(2005) - et al.
Sexual dimorphism of brain developmental trajectories during childhood and adolescence
NeuroImage
(2007)
Human cerebral cortex: a system for the integration of volume- and surface-based representations
NeuroImage
Thinking of the future and past: the roles of the frontal pole and the medial temporal lobes
NeuroImage
Conflict-related activity in the caudal anterior cingulate cortex in the absence of awareness
Biol. Psychol.
Evidence for abnormalities of cortical development in adolescent-onset schizophrenia
NeuroImage
Improved prediction of response to antiretroviral combination therapy using the genetic barrier to drug resistance
Antivir. Ther.
Which classification algorithm works best with stylistic features of Portuguese in order to classify web texts according to users needs? Technical Report NILC-TR-04-09
MRI volumes of amygdala and hippocampus in non-mentally retarded autistic adolescents and adults
Neurology
Effects of age on brain volume and head circumference in autism
Neurology
The cortical signature of prodromal AD: regional thinning predicts mild AD dementia
Neurology
Autism as a disorder of neural information processing: directions for research and targets for therapy
Mol. Psychiatry
Controlling the false discovery rate: a practical and powerful approach to multiple testing
J. R. Stat. Soc., Series B (Methodological)
The precuneus: a review of its functional anatomy and behavioural correlates
Brain
Association study between BDNF gene polymorphisms and autism by three-dimensional gel-based microarray
Int. J. Mol. Sci.
Unusual brain growth patterns in early life in patients with autistic disorder: an MRI study
Neurology
The cortical signature of Alzheimer's disease: regionally specific cortical thinning relates to symptom severity in very mild to mild AD dementia and is detectable in asymptomatic amyloid-positive individuals 10.1093/cercor/bhn113
Cereb. Cortex
Investigating the predictive value of whole-brain structural MR scans in autism: a pattern classification approach
NeuroImage
Measuring the thickness of the human cerebral cortex from magnetic resonance images
Proc. Natl. Acad. Sci. U. S. A.
Automatically parcellating the human cerebral cortex
Cereb. Cortex
Cited by (204)
An exploration of structural brain differences in Autism Spectrum Disorders: A multi-parcellation and multi-age analysis
2024, Biomedical Signal Processing and ControlHybrid cuttle Fish-Grey wolf optimization tuned weighted ensemble classifier for Alzheimer's disease classification
2024, Biomedical Signal Processing and ControlSyndrome-Specific Neuroanatomical Phenotypes in Girls With Turner and Noonan Syndromes
2024, Biological Psychiatry: Cognitive Neuroscience and NeuroimagingRevealing the Selectivity of Neuroanatomical Alteration in Autism Spectrum Disorder via Reverse Inference
2023, Biological Psychiatry: Cognitive Neuroscience and NeuroimagingAutism spectrum disorder diagnosis using the relational graph attention network
2023, Biomedical Signal Processing and ControlInsulin and disorders of behavioural flexibility
2023, Neuroscience and Biobehavioral Reviews