Robin’s Viewer: Using Deep-Learning Predictions to Assist EEG Annotation

Abstract

Deep learning has been successfully applied to EEG data for sleep staging, epileptic seizure detection, and artifact recognition. However, the performance of automated annotation is not yet sufficient to replace trained annotators in a clinical setting. Therefore, we propose a decision-support-system to help human annotators work faster and more effectively. As a first step towards addressing these challenges, we developed Robin's Viewer (RV) in Python, building on top of the plotting library Plotly and the popular M/EEG analysis toolbox MNE. The goal was to create a platform-independent, interactive web app, which is open-source and supports many common EEG-file formats to facilitate easy integration with a variety of EEG toolboxes. RV includes many common features of other EEG viewers, e.g., a view-slider, marking of bad channels and transient artifacts, and customizable preprocessing. The key feature distinguishing RV from existing EEG viewers is the possibility to visualize output predictions of deep-learning models that are trained to recognize patterns in EEG data. The result is a decision-support-system for scientists and clinicians who can use RV to annotate artifacts, sleep stages, abnormalities, and other classification tasks.

Improved Manual Annotation of EEG Signals through Convolutional Neural Network Guidance

“…This may be a way of evaluating upstream cleaning and leaving room for further cleaning without having to manually inspect a large number of segments already properly cleaned by these algorithms. By integrating our method into a signal viewer that is currently being designed in our group ( Weiler et al, 2022 ), we expect to facilitate fast and reliable resting-state EEG data analyses.…”