Statistical Inference for Wavelet-denoised Statistical Parametric Maps
Casanova R, Srikanth R, Laurienti P, Peiffer A, Maldjian J
Wake Forest University School of Medicine
We present a novel method for threshoding wavelet-denoised statistical parametric maps. Current wavelet-denoising methods produce smoothed statistical maps that then require another thresholding step in the spatial domain. The choice of this threshold has thus far been arbitrary. In our proposed framework, a rejection region is fixed for the wavelet-denoised maps, and the achieved false discovery rate (FDR) is estimated. Various techniques were used to assess the FDR control and their performance was evaluated using both simulated and in-vivo fMRI data. The methods accounting for spatial correlation of SPMs performed better than methods that do not.