Functional MRI analysis using training-based prior models of activation patterns.

摘要

Functional Magnetic Resonance Imaging (fMRI) has become a powerful tool for research into human brain function due to its non-invasiveness and high resolution. However, the strength of the observed functional signal is small and contaminated by systematic and physiological noise, presenting a formidable challenge to analysis. Although strategies that exploit a priori knowledge of functional activations have been demonstrated to be more robust, prior information in existing analytical techniques has been limited to the spatially contiguous and locally homogenous nature of fMRI responses.;Drawing inspiration from the success enjoyed by statistical shape priors in the field of image segmentation, this work introduces statistical activation priors to fMRI data analysis. This novel idea involves learning brain activation patterns strength, shape and location) from previously conducted multi-subject fMRI studies (training data) to define functionally informed priors for improved analysis of new subjects. A Bayesian framework is used to incorporate prior information through prior probability distributions and to find the maximum a posteriori (MAP) estimate of activation strength parameters. Both principal component analysis (PCA) and independent component analysis (ICA) are investigated for capturing variation in activation patterns across training samples. To establish the importance of training-based functional prior information in combating low signal-to-noise ratio (SNR), spatial priors that utilize only local regularization and anatomical constraints are also evaluated.;Results from synthetic as well as real fMRI data illustrate that training-based priors provide statistically significant improvement in estimating activation compared with standard general linear model (GLM) based methods. Unlike other spatial regularization priors, functional activation priors compensate for low SNR by inducing sensitivity to task-related regions of the brain. Consequently, a major advantage of statistical activation priors is their potential to reduce the amount of time series data needed and the corresponding image acquisition time in test subjects. The sensitivity of the proposed estimation methods to model parameters and the application of different estimators to multi-group fMRI studies is also investigated. Both single and multi-group fMRI experiments establish that ICA-based priors are better equipped to handle inter-subject variability in functional anatomy than their PCA counterparts.