A novel t-test for low-SNR fMRI brain mapping

Detecting signal in fMRI studies relies on the classical testing framework developed for Gaussian signals. Unfortunately, fMRI signals are amplitude measurements such that the signal follows a Rice distribution. The classical t-test used for detection performs reasonably well for signals with a high Signal-to-Noise Ratio (SNR). To accurately detect the voxels at the border of the brain region of interest, we need to deal with small SNRs such that dealing with the Rice distribution is a necessity. Most techniques dealing with the Rice distribution require amplitude estimates based on the Maximum Likelihood Estimate (MLE) requiring an iterative approach and may still lead to a false local solution. The analytical alternative to the MLE is by applying the Method-of-Moment (MoM) estimator which performs better for low SNR conditions than the Gaussian framework. In this paper, we propose a novel t-test based on the MoM-estimates for signal and noise power to assess voxel activity and to generate the according brain map.