Robust Hemodynamic Response Function Estimation From fNIRS Signals

Hemodynamic response function (HRF) estimation in functional near-infrared spectroscopy (fNIRS) plays an important role in characterizing the temporal dynamics of the brain response. Estimation based on semiparametric modeling is very useful for fNIRS signals. However, the Gaussian noise assumption may be too simplistic since the sources of noise and their nature are various in fNIRS. In this paper, the problem of HRF estimation using a semiparametric model whose nonparametric part is viewed as a nuisance component used to represent the drift is considered. A new robust HRF estimation procedure to minimize the impact of unexpected noise in fNIRS signals is developed. Within the proposed estimation method, the drift effect is removed by applying a first order difference to the fNIRS signal samples. Consistency and asymptotic normality of the proposed estimator are established and its effectiveness is illustrated on both simulated and real fNIRS data.