An Application of Regularized Spectral Entropy for Detection of Task-Related Information Content in fMRI

Abstract

Functional MRI (fMRI) has become a critical tool for clinical evaluation and neuroimaging research in recent years. This work investigates a unique application of spectral entropy (Shannon’s entropy in the frequency domain) combined with a regularization scheme for an ill-fitted problem to identify the presence of useful task-related information content in fMRI scans. Regularized spectral entropy was compared to traditional methods of identifying useful information such as the General Linear Model (GLM), as well as known percent signal change in simulated data sets created with noise parameters informed by real data sets, and signal-to-noise ratio (SNR) in idealized signals. Combined with regularization, spectral entropy was found to have comparable sensitivity and specificity to the GLM, as well as a correlated response to percent signal change and SNR. Additionally, spectral entropy was fast to compute and required minimal a priori information compared to other methods used to identify useful task-related information.