Autism is a complex developmental disability, characterized by deficits in social interaction, communication skills, range of interests, and occasionally the presence of stereotyped behaviors. Several studies show that changes in brain weight and volume over aging follow a unique trajectory in those affected by the condition (Redcay E, Courchesne E. When is the brain enlarged in autism? A meta-analysis of all brain size reports. Biol Psychiatry 58(1):1-9, 2005). In this work, we develop a robust technique for evaluating the volume of white matter (WM), and use it as the main classification criteria. We perform MRI-based analysis on the brains of 14 autistic and 28 control subjects, male and female between aged 7 to 38 years. The proposed framework consists of several stages. First the entire T1-weighted MRI scans are filtered out from noise using anisotropic diffusion filter. Then, the white matter (WM) is segmented from the skull. The segmentation framework is the search for maximum-a-posterior configurations in a Markov Gibbs Random Field (MGRF) model. After that, a 3D mesh is generated from the segmented WM. Finally, the volume of the 3D mesh is computed using a new algorithm. The experiments show accurate classification results of the proposed framework.

The maximum a posteriori (MAP) technique, combining intensity and spatial interactions, has been a standard statistical approach for image segmentation. Crucial steps for the MAP technique are the model identification, incorporation of priors, and the optimization approach. This paper describes an unsupervised MAP segmentation framework of N-dimensional multimodal images. The input image and its desired labelling are described by a joint Markov-Gibbs random field (MGRF) model of independent image signals and interdependent region labels. A kernel approach is used to model the joint and marginal probability densities of objects from the gray level histogram, incorporating a generalized linear combination of Gaussians (LCG). A novel maximum likelihood estimate (MLE) for the number of classes in the LCG model is introduced. An approach is devised for MGRF model identification based on region characteristics. The segmentation process employs LCG to provide an initial segmentation, and then alpha-expansion move algorithm iteratively refines the labelled image using MGRF. The resulting MAP algorithm is studied in terms of convergence and sensitivity to initialization, improper estimation of the number of classes, and discontinuities in the objects. The framework is modular, allowing incorporation of intensity and spatial interactions with varying complexity, and can be extended to incorporate shape priors.