Automatic 1D convolutional neural network-based detection of artifacts in MEG acquired without electrooculography or electrocardiography

Abstract

Magnetoencephalography (MEG) is a functional neuroimaging tool that records the magnetic fields induced by electrical neuronal activity; however, signal from non-neuronal sources can corrupt the data. Eye-Blinks (EB) and Cardiac Activity (CA) are two of the most common types of non-neuronal artifacts. They can be measured by affixing eye proximal electrodes, as in electrooculography (EOG) and chest electrodes, as in electrocardiography (EKG), however this complicates imaging setup, decreases patient comfort, and often induces further artifacts from facial twitching and postural muscle movement. We propose an EOG- and EKG-free approach to identify eye-blink, cardiac, or neuronal signals for automated artifact suppression. Our contributions are two-fold. First, we combine a data driven, multivariate decomposition approach based on Independent Component Analysis (ICA) and a highly accurate classifier constructed as a deep 1-D Convolutional Neural Network. Second, we visualize the features learned to reveal what features the model uses and to bolster user confidence in our model’s training and potential for generalization. We train and test three variants of our method on resting state MEG data from 49 subjects. Our cardiac model achieves a 96% sensitivity and 99% specificity on the set-aside test-set. Our eye-blink model achieves a sensitivity of 85% and specificity of 97%. This work facilitates automated MEG processing for both, clinical and research use, and can obviate the need for EOG or EKG electrodes.

Deep Learning Approach for Automatic Classification of Ocular and Cardiac Artifacts in MEG Data

“…The results obtained from our model are comparable to recently published work of [44][45][46]. However, we have demonstrated that when temporal and spatial information are combined into one model the artifact rejection performance is increased (Figure 7).…”

“…In their work, the neural networks are trained to classify ECG and EOG related artifact components from the timeseries of the ICs. In a related study, the same authors showed automatic detection of EOG related components using a spatial representation of the independent components [44]. As a part of their conclusions they suggest a neural network which combines both temporal and spatial information, which is ideal.…”

“…The method has been shown to be quite robust when dealing with EEG data, but translation to MEG maybe more difficult and time consuming. More recently, Garg et al presented a method using multilayered CNNs to classify artifact components from MEG data [44,45]. In their work, the neural networks are trained to classify ECG and EOG related artifact components from the timeseries of the ICs.…”

“…In [48] labeling was performed by experts making decisions from visual inspection of the ICA decomposed signals. In contrast, we used statistical measures to identify the set of reference components [20,32,44,45]. This choice of labeling cannot be considered as the ground truth either.…”

Decreased Task-Related HRV Is Associated With Inhibitory Dysfunction Through Functional Inter-Region Connectivity of PFC in Major Depressive Disorder

“…To guarantee HRV could reflect task-related state more precisely, ECG record was directly collected from the MEG raw data under the go/no-go task after ICA (36). There were two reasons to choose time domain analysis.…”