Swarm Intelligence-Based Improved Adaptive Chirp Mode Decomposition Algorithm for Suppression of Ocular Artifacts from EEG Signal

• Published in: IEEE sensors journal Vol. 24; no. 6; pp. 8314 - 8325
• Main Authors: Silpa, Bommala, Hota, Malaya Kumar
• Format: Journal Article
• Language: English
• Published: New York IEEE 15.03.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Adaptive chirp mode decomposition (ACMD); Algorithms; Chirp; Clustering; Continuous wavelet transform; continuous wavelet transform (CWT); Continuous wavelet transforms; Decomposition; electroencephalogram (EEG); Electroencephalography; Energy value; Eye movements; k-means clustering algorithm; Mathematical analysis; Neurological diseases; Noise measurement; ocular artifacts (OAs); Optimization; Recording; Subtraction; Swarm intelligence; Transforms; Waveforms; Wavelet transforms
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2024.3356579

Electroencephalogram (EEG) signals are mostly contaminated with ocular artifacts (OAs) due to eye movements and eye blinks. These artifacts make the EEG recordings difficult to analyze and diagnose neurological diseases. Therefore, in this work, we propose a hybrid framework that consists of two stages to remove OAs from the EEG signals. A correlation-based variational mode decomposition (VMD) method in the first stage removes baseline wander (BW) noise. Then, an improved adaptive chirp mode decomposition (IACMD) with continuous wavelet transform (CWT) and <inline-formula> <tex-math notation="LaTeX">{k} </tex-math></inline-formula>-means clustering algorithm in the second stage detects and removes OAs. The IACMD, in the second stage, is implemented by optimizing ACMD parameters using an improved grey wolf optimization (IGWO) algorithm with a correlation waveform index (Cwi) as a fitness function. The IACMD extracts the modes, and the noisy mode is identified based on the energy value. However, the direct subtraction of the identified noisy mode may eliminate some information from the EEG signal. Hence, in this work, CWT and <inline-formula> <tex-math notation="LaTeX">{k} </tex-math></inline-formula>-means clustering algorithms are used to estimate the OA-affected interval by separating EEG elements from the noisy mode. Finally, the denoised EEG signal is determined by subtracting the estimated OA signal without impacting the non-artifactual regions. The analysis is carried out on the MIT-BIH polysomnographic and EEG during mental arithmetic tasks (EEGMAT) databases. Compared to the existing techniques, the proposed method shows superior performance in terms of efficiency and preservation of EEG data. Further, the subject-wise and length-wise analysis reveals the robustness of the proposed framework.