Reliability Assessment of Denoising Methods for Change Point and Activation Detection in Lack of Ground Truth from fMRI

This study presents and discusses the outcomes of different de-noising approaches, which are applied to low SNR real fMRI BOLD signals. These signals are acquired during a psychiatric fMRI experiment that is specifically designed for observing emotional mood changes (EMC). Despite a priori knowledge about activations and change points for motor tasks, EMC signals in this study do not have a ground truth. Up to our knowledge, the challenges of such a task has never been visited. Accordingly, this study aims to compare the performance of three de-noising methods, which have inherently different characteristics, on preserving the waveform characteristics of change points and activation durations while removing noise from EMC-BOLD signals through expert guided qualitative analysis.