NAVIGATING NOISE IN RODENT EEG

Spike-wave discharges (SWDs) are high amplitude, short oscillatory bursts of neuronal activity associated with absence seizures, recorded in electroencephalography (EEG). They are critical to detect in rodent EEG data, to ensure that neural signals of interest are not contaminated, and to localize and quantify them for further analysis. Detecting and removing noise from EEG data, particularly in the context of SWDs, remains a significant challenge. Manual processing is labor-intensive, while existing automated methods often fail to generalize across datasets due to variations in experimental conditions, equipment, and animal strains. To address this, we developed a flexible analysis pipeline implemented as a MATLAB application, designed to adapt to diverse datasets and streamline the identification and labeling of noisy events in the data, including SWDs. The application generates a dimensionally reduced representation of the data segments, using a combination of features based on signal statistics, power spectral bands and parametric SWD detection algorithms from literature. The application allows the users to visualize data segments as clusters of interest, inspect raw data within clusters, and iteratively adjust and label the cluster selections. This approach enables efficient labeling of large datasets, facilitating the detection and removal of dataset-specific noise with minimal manual effort. Additionally, the broad set of features from data segments and their low dimensional representation allows the application to handle various noise types beyond SWDs. We aim to extend the application to incorporate active learning framework for improved detection accuracy with minimal manual intervention.