Wavelet-based denoising and independent component analysis for improving multi-group inference in fMRI data

Denoising is amongst the most challenging steps involved in analyzing fMRI data. The conventionally used Gaussian smoothing improves the SNR at the cost of spatial sensitivity and specificity. We briefly describe a 3-D framework for wavelet based fMRI analysis that includes denoising and signal separation followed by a detailed illustration of the benefits and improvements when applied to multi-group (healthy/patient) fMRI studies. We utilize a novel shape metric to highlight the accuracy of the shape of activation regions obtained through different processing frameworks. The proposed algorithm results in higher specificity and increased shape accuracy which in turn is likely to be more sensitive to important differences in the patient and control group.