Concurrent bias correction in hemodynamic data assimilation.

Area of Science:

• Neuroimaging
• Signal Processing
• Biomedical Engineering

Background:

• Low-frequency drift in fMRI data complicates assimilation by introducing uncertainties.
• Conventional methods struggle with the unmodeled nature of drift in fMRI.

Purpose of the Study:

• To develop a novel approach for assimilating nonlinear hemodynamic systems with stochastic biased noise, specifically addressing low-frequency drift in fMRI.
• To improve the accuracy and robustness of fMRI data analysis by effectively handling time-varying bias and drift.

Main Methods:

• Treated drift variation as a random-walk process, transforming the assimilation problem into nonlinear system identification with time-varying bias.
• Developed a bias aware unscented Kalman estimator with two parallel filters for estimating bias-free states and drift separately.
• Corrected system state estimates using bias-free states and drift estimates in an online fashion.

Main Results:

• The proposed method successfully assimilates fMRI responses and drift simultaneously without assumptions on drift characteristics.
• Experiments with synthetic and real fMRI data demonstrated the feasibility and advantages over classic polynomial approaches.
• The approach effectively handles time-varying bias and drift, crucial for accurate fMRI analysis.

Conclusions:

• The bias aware unscented Kalman estimator provides a robust and flexible framework for fMRI data assimilation in the presence of drift.
• This method offers significant advantages for fMRI imaging where drift modeling is challenging.
• The approach enhances the reliability of fMRI analysis by accurately separating neural signals from scanner-induced drift.