Tensor decomposition as a useful tool for eye blink removal in EEG

Investigators: Roman Rosipal, Zuzana Rošťáková

 An electroencephalogram (EEG) recorded with eyes open is contaminated by artifacts related to eye blinks and eye movements, which should be removed before further analysis. For this purpose, we developed SPECTER, an algorithm that combines tensor decomposition of the EEG spectrum, removal of artifact components, and spectrum-to-signal transformation. In contrast to traditional approaches, SPECTER does not require information about the electrooculogram (EOG), which simplifies its application. We also analyzed the effect of component removal and identified approaches that may contaminate the signal with new, artificial artifacts – a common problem in independent component analysis (ICA). However, SPECTER uses simple component subtraction, minimizing the risk of data contamination, and includes criteria for identifying problematic cases. Real data analysis has proven that SPECTER produces comparable or superior results compared to traditional approaches and represents a promising tool for EEG signal preprocessing.

Projects: 09I03-03-V04-00205, 09I03-03-V04-00443, APVV-21-0105

Fig. 1: Graphical scheme of the SPECTER algorithm which was developed for eye blink detection and removal from the EEG signal. The EEG signal corrupted by eye blinks is depicted in grey, while its cleaned version is depicted in black.

Publications:

  • Rošťáková, Z. – Rosipal, R. – Trejo, L.J. SPECTER – The Signal sPECtrum Tensor decomposition and Eye blink Removal algorithm. In Biomedical Signal Processing and Control, 2025, vol. 99, art. no. 106889. (2024: 4.9 – IF, Q2 – JCR, 1.229 – SJR, Q1 – SJR). ISSN 1746-8094. Available on: https://doi.org/10.1016/j.bspc.2024.106889
  • Rošťáková, Z. – Rosipal, R. Deflation properties in tensor-based eye blink removal algorithm. In Statistical Papers, 2025, vol. 66, art. no. 91. (2024: 1.1 – IF, Q3 – JCR, 0.775 – SJR, Q2 – SJR). ISSN 0932-5026. Available on: https://doi.org/10.1007/s00362-025-01714-w
  • Rošťáková, Z. – Rosipal, R. Introducing SPECTER 2.0 – an enhanced version of the tensor based eye blink removal algorithm. In Proceedings of the 15th International Conference on Measurement. – Bratislava, Slovakia : Institute of Measurement Science, SAS, 2025, p. 53-56. ISBN 978-80-69159-01-3. Available on: https://doi.org/10.23919/MEASUREMENT66999.2025.11078683