Analysis of Imaging Patterns Using Pattern Recognition Methods: Application to Development of Imaging-Based Biomarkers for Autism

This talk reviews relatively state of the art image analysis methods for understanding imaging patterns that reflect underlying structural and functional differences between autistic and TDC children. Special emphasis is given on pattern analysis and classification  methods, which utilize nonlinear multivariate machine learning principles to determine distinctive imaging phenotypes/biomarkers that can be used for classification of individuals. The talk will discuss conventional measurements of brain volumes and regions of interest, it will then present new opportunities provided voxel- and pattern-based morphometry. The talk will then discuss pattern classification methods, their potential emanating from their ability to measure complex and subtle imaging phenotypes, and their pitfalls related to their high dimensionality and the frequently improper use or evaluation of these methods. Examples from other neuropsychiatric and neurologic disorders, and particularly schizophrenia and AD, in which these methods have been used more extensively, will also be used to demonstrate the potential of these methods, especially from the perspective of predictive imaging biomarkers. Finally, we will discuss the potential of jointly using functional, structural and connectivity imaging biomarkers to achieve higher classification accuracy.