State-space modeling based on principal component analysis and oxygenated-deoxygenated correlation to improve near-infrared spectroscopy signals

Abstract

Near infrared spectroscopy (NIRS) is currently becoming an effective technique for noninvasive functional brain imaging. Therefore, the methods to improve the quality of measured NIRS signals play an important role to make NIRS broadly accepted in practical applications. Previously, there have been approaches using state-space modeling to recover the NIRS signals from basic component signals to eliminate the artifacts presented in the NIRS measurements. However, the proposed approach requires us an onset vector to determine the starting position of stimulus that is not always available in practical situation. In this work, we provide a new way to find the basic components for efficient implementations of the state-space modeling. We apply principal component analysis to estimate eigenvector-based basis that presents the compact information of the whole signals. We utilize the oxygenated-deoxygenated correlation to find another set of basic components to enhance the quality of NIRS signals. The state-space modeling based on Kalman filter is used to reconstruct the NIRS signals from these basic components. We tested the proposed algorithm with actual data and showed significant improvements of the contrast-to-noise (CNR) of the NIRS signals after filtered by our proposed approach.