Wavelet-based Artifact Removal Algorithm for EEG Data by Optimizing Mother Wavelet and Threshold Parameters

EEG recordings are usually affected by various artifact types come from non-neural sources and make it difficult for accurate signal classification in the later stage. Thus reliably detecting and removing artifacts from EEG by an automated signal processing algorithm is an active research area. In this paper we have developed a wavelet based artifact removal algorithm from EEG data that selects the best (optimal) threshold parameters, and hence consequently provides the best performance of artifact removal. In the proposed algorithm we choose to sweep both the wavelet filter parameter and threshold parameters until the best accuracy and/or least distortion is achieved by making a decision based on a reference dataset. The criteria for optimized selection are based on the metrics that quantify both amount of artifact removal and amount of distortion in the signal in both time and frequency domain. The algorithm is tested on synthesized EEG data that include different artifact templates and thus quantifies the performance based on several time and frequency domain measures. The achieved results prove that by selecting the optimum mother wavelet and parameter values adaptively would give the best performance both with regard to amount of artifact removal and least signal distortion compared with selecting any predefined mother wavelet and/or constant threshold parameter. This research would help the EEG signal analysis community a platform to work further in future on such problem to be able to properly select the wavelet parameters.