A Novel Deep Learning Denoising Algorithm for Neural Signal Recovery in fMRI Scanning

Motivation: While fMRI infers neural activity from hemodynamic changes, the relationship between the two remains to be further clarified. Simultaneous electrophysiological recordings (Ephy) and fMRI can provide additional insights into neurovascular coupling and brain function. Goal(s): Our objective is to address the electromagnetic interference (EMI) noise in the simultaneous Ephy and fMRI recording. Approach: A deep learning-based fully convolutional neural network (FCNN) was proposed to effectively eliminate EMI noise. Simulated neural signals and tactile-evoked neural signals were implemented for training and testing. Results: FCNN significantly reducing EMI noises, maintaining spike waveform consistency and successfully retaining the most neural signals. Impact: This research proposed a universal and robust denoising approach to address electromagnetic interference during simultaneous recording of neural signals and fMRI data, which will be relevant for understanding of neurovascular coupling and brain function.