The Riemannian Potato Field: A Tool for Online Signal Quality Index of EEG

Abstract

Electroencephalographic (EEG) recordings are contaminated by instrumental, environmental, and biological artifacts, resulting in low signal-to-noise ratio. Artifact detection is a critical task for real-time applications where the signal is used to give a continuous feedback to the user. In these applications, it is therefore necessary to estimate online a signal quality index (SQI) in order to stop the feedback when the signal quality is unacceptable. In this article we introduce the Riemannian Potato Field (RPF) algorithm as such SQI. It is a generalization and extension of the Riemannian Potato, a previously published real-time artifact detection algorithm, whose performance is degraded as the number of channels increases. The RPF overcomes this limitation by combining the outputs of several smaller potatoes into a unique SQI resulting in a higher sensitivity and specificity, regardless of the number of electrodes. We demonstrate these results on a clinical dataset totalizing more than 2200 hours of EEG recorded at home, that is, in a non-controlled environment.

A Riemannian Modification of Artifact Subspace Reconstruction for EEG Artifact Handling

“…Since the space of SPD matrices is a (negatively) curved space , the use of traditional Euclidean geometry, which implies that distances are computed along straight lines in the data space, turns out to be disadvantageous. Riemannian geometry has shown to be more efficient and yields more precise results in the analysis of EEG data (Shyu et al, 2006; Kalunga et al, 2015; Yger et al, 2015; Congedo et al, 2017; Horev et al, 2017; Barthélemy et al, 2019).…”

“…In rASR, a sample covariance matrix is computed as a robust, unbiased estimator of the covariance matrix of the current data segment (Kalunga et al, 2015). It is defined as: normalU=1t-1⁢XXnormalT, where t is the number of samples in the current data segment and X ∈ ℝ t×c is the current channel matrix consisting of t samples and c channels (Barachant et al, 2012; Barthélemy et al, 2019). By computing an estimator of the covariance matrix of the current data segment held in the method, the rASR method omits the necessity of computing individual covariance matrices for every small chunk of the data defined by stepsize, as described in the ASR paragraph.…”

“…Due to their mathematical properties though, covariance matrices should be processed using methods applicable for curved, high-dimensional data spaces (Moakher and Batchelor, 2006; Yger et al, 2015; Lotte et al, 2018). For such computations, Riemannian geometry has been shown to be more precise than Euclidean methods (Barachant et al, 2012, 2013; Kalunga et al, 2015; Yger et al, 2015; Horev et al, 2017; Lotte et al, 2018) and has been used in online artifact detection and correction previously (Barthélemy et al, 2017, 2019). Therefore, we speculated that a Riemannian-adapted ASR could outperform the original ASR algorithm.…”

SMR/Theta Neurofeedback Training Improves Cognitive Performance and EEG Activity in Elderly With Mild Cognitive Impairment: A Pilot Study

“…We have removed the artifact using independent component analysis (ICA) decomposition (Gramfort et al, 2014), removing independent components linked with ocular movements and muscular artifacts. The recorded EEG also passed a rejection test with a covariance-based approach (Barthelemy et al, 2019), ensuring that there are no remaining artifacts. All the ECB conditions were concatenated to estimate the variation in the frequency domain, relying on the Welch method.…”

Is there a cluster of high theta-beta ratio patients in attention deficit hyperactivity disorder?

“…iAPF) was extracted for the computation of individualized frequency bands, and the TBR was calculated as described below. The method presented in this paper to calculate and compare TBR values on eyes open introduces steps to make the estimates of the TBR index more reliable.First, artifactual data is removed from the original time series using the Riemannian Potato Field (RPF)(Barthélemy et al, 2019). EEG recordings were epoched (2 seconds length, overlapped every 0.125 second) and the covariance matrix of each epoch is computed for a subset of channels.…”