Wavelet-based fMRI analysis: 3-D denoising, signal separation, and validation metrics

►A novel 3-D denoising algorithm that utilizes 3-D wavelet transform and signal estimation theory to compute the denoised estimates from noisy task-related simulated fMRI data (SNR ~ 11-14 dB). ►Utilize these denoised estimates for estimating the activation maps using independent component analysis in the wavelet domain. Performing ICA in the wavelet domain utilizes the de-correlated nature of wavelet coefficients to extract maximally independent components. ►Introduce two new metrics for validation of the shape (and spatial extent) of the activation regions by utilizing their geometric properties such as perimeter and Centroids. ►The above framework was tested on large of signal and noise levels (for simulated data) in addition to a real motor-tapping task data set. Results show that the proposed methodology and wavelets in general are able to achieve superior accuracy as compared to the conventionally used framework. ►The main advantage of our framework is its capability to accurately extract the shape of an activation region after large amount of processing. This feature is expected to help researchers performing studies on large number of subjects in control or patient groups to obtain more accurate activation maps.