A methodology to improve estimation of stimulus-evoked hemodynamic response from fNIRS measurements