Abstract #3592

Dynamical Statistical Modeling of Physiological Noise for Fast BOLD fMRI

Simo Sarkka¹, Aapo Nummenmaa^1,2, Arno Solin¹, Aki Vehtari¹, Thomas Witzel³, Toni Auranen⁴, Simo Vanni⁴, Matti S. Hamalainen², Fa-Hsuan Lin^1,5

¹Department of Biomedical Engineering & Computational Science, Aalto University, Espoo, Finland; ²Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, United States; ³Harvard-MIT Division of Health Sciences & Technology, Harvard University, Cambridge, MA, United States; ⁴Advanced Magnetic Imaging Centre, Low Temperature Laboratory, Aalto University, Espoo, Finland; ⁵Institute of Biomedical Engineering, National Taiwan University, Taipei, Taiwan

In this work, we propose a statistical model based method for removal and analysis of physiological noise in fast BOLD fMRI acquisition methods. The proposed stochastic state space model allows for accurate dynamic tracking of time-varying physiological signal frequencies and the estimation method is based on the Interacting Multiple Models (IMM) Kalman filter (KF) algorithm, which is widely used in real time target tracking applications. The method forms statistically the best possible separation of the spatiotemporal BOLD and physiological signals into separate components, which allows for further processing of the de-noised BOLD signal or analysis of the spatial characteristics of the physiological signals. The proposed method was applied to a three-slice EPI data and the results indicate that the method is able to accurately separate the cardiac and respiration signals from the BOLD signal.