Piecewise empirical mode Bayesian estimation – A new method to denoise electrooculograms. (September 2021)

• Record Type: Journal Article
• Authors: Dasgupta, Anirban; Routray, Aurobinda
• Is Part Of: Biomedical signal processing and control. Volume 70(2021)
• Publication Date: 2021-09
• Subjects: EOG -- Baseline wander -- Quadratic variation -- Denoising -- Empirical mode decomposition; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102945
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: The aim of the paper is to denoise EOG signals, while preserving the signatures in regions of changes in the type of eye movements. The signal into sub-signals based on the region of changes in the eye movement type, followed by empirical mode decomposition of each sub-signal, concatenation of the reconstructed sub-signals, and finally Bayesian estimation to remove reconstruction artifacts. We name this algorithm as the Piecewise Empirical Mode Bayesian Estimation (PEMBE). PEMBE is found to be more effective in the denoising process as compared to the existing methods, as tested on the Physiosig EOG. Abstract: Electrooculograms (EOG) indicate the magnitude and direction of eye movements. The EOG contain several types of noise and interference, which makes their analysis difficult. The existing denoising methods fail to preserve certain eye movement signatures, mainly in regions where there is a change of eye movement type, for e.g., fixation to saccades. Hence, the prime aim of the study is to denoise EOG signals, while preserving the signatures in such conditions. In this paper, we propose a new denoising method that addresses the aforementioned issue. In this algorithm, we first divide the signal into sub-signals based on the region of changes in the eye movement type. For each sub-signal, we separate the intrinsic mode oscillations from noise. Finally, the filtered sub-signals are concatenated to form the reconstructed signal. The reconstruction artifacts at the joining region are removed by Bayesian estimation. We name this algorithm as the Piecewise Empirical Mode Bayesian Estimation (PEMBE). We demonstrate that PEMBE is more effective in the denoising process as compared to the existing methods, as tested on the Physiosig EOG. The PEMBE is also found to be useful in the removal of the baseline wander.