A framework for fully automated and online artefact denoising in single-channel electroencephalography using wavelet transform and adaptive unsupervised learning

Abstract

Electroencephalography is a technique that records brain activity from the scalp with high temporal resolution and in a non-invasive manner. Electroencephalographic signals (EEG) are highly sensitive recordings that are inherently prone to contamination from artefacts generated by muscular activity, eye movements, power-line interference, and electrode displacement, among others. Such forms of noise can distort the neural patterns of interest, thereby hampering subsequent analyses of cognitive states and influencing diagnoses of neurological disorders, as well as affecting the control of Brain-Computer Interface (BCI) systems. While many methods have been devised and implemented, most are offline. This means they operate only after an entire EEG recording is completed. However, in many practical applications, including closed-loop BCIs, seizure detection, neurofeedback, and brain-controlled prosthetics, EEG signals must be recorded, processed, and interpreted in real time. This is because essential decisions must be made promptly for safety reasons, to support therapeutic effectiveness, or to enhance the user experience. Effective online artefact removal is imperative if the performance of an adaptive system must be guaranteed, and actions cannot be delayed.