EEMD and bidimensional RLS to suppress physiological interference for heterogeneous distribution in fNIRS study

Abstract

Near-infrared spectroscopy (NIRS) can provide the hemodynamics information based on the hemoglobin concentration representing the blood oxygen metabolism of the cerebral cortical, which can be deployed for the cerebral function study. However, NIRS-based cerebral function detection accuracy can be significantly influenced by the physiological activities such as cardic cycle, respiration, spontaneous low-frequency oscillation and ultra-low frequency oscillation. The distribution difference of the capillary, artery and vein leads to the heterogeneity feature of the cerebral tissues. In the case that the heterogeneity is not serious, good detection accuracy and stable performance can be achieved through the regression analysis as the reference signal can well represent the interference in the measurement signal when conducting the multi-distance measurement approach. The direct use of the reference signal to estimate the interference is not able to achieve good performance in the case that the heterogeneity is serious. In this study, the cerebral function activity signal is extracted using recursive least square (RLS) method based on the multi-distance measurement method in which the reference signal is processed by ensemble empirical mode decomposition (EEMD) algorithm. The temporal and dimensional correlation of the neighboring sampling values are applied to estimate the interference in the measurement signal. Monte Carlo simulation based on a heterogeneous model is adopted here to investigate the effectiveness of this methodology. The results show that this methodology can effectively suppress the physiological interference and improve the detection accuracy of cerebral activity signal.