|
Abstract:
|
This talk introduces new methods for analyzing complex image data (e.g., 2d gel proteomics, brain fMRI). It involves extending the functional mixed model to image data, relating the image to covariates through "fixed effect images" indicating the effects at various locations in the image. The inference accounts for correlation between images using random effect structures, and the wavelet-based modeling approach accommodates nonstationary covariance features within the image, enabling adaptive smoothing of the images and adaptive borrowing of strength between pixels in resulting inference. The automatic, efficient Bayesian model fitting algorithm can be feasibly fit to enormous data sets, and yields posterior samples for various Bayesian inferences, including the determination of regions of the image related to a given covariate of interest while controlling the false discovery rate.
|
