Quantifying Signal Quality From Unimodal and Multimodal Sources: Application to EEG With Ocular and Motion Artifacts.

With prevalence of electrophysiological data collected outside of the laboratory from portable, non-invasive modalities growing at a rapid rate, the quality of these recorded data, if not adequate, could affect the effectiveness of medical devices that depend of them. In this work, we propose novel methods to evaluate electrophysiological signal quality to determine how much of the data represents the physiological source of interest. Data driven models are investigated through Bayesian decision and deep learning-based methods to score unimodal (signal and noise recorded on same device) and multimodal (signal and noise each recorded from different devices) data, respectively. We validate these methods and models on three electroencephalography (EEG) data sets ( = 60 subjects) to score EEG quality based on the presence of ocular artifacts with our unimodal method and motion artifacts with our multimodal method. Further, we apply our unimodal source method to compare the performance of two different artifact removal algorithms. Our results show we are able to effectively score EEG data using both methods and apply our method to evaluate the performance of other artifact removal algorithms that target ocular artifacts. Methods developed and validated here can be used to assess data quality and evaluate the effectiveness of certain noise-reduction algorithms.