Joint maximum likelihood estimation of activation and Hemodynamic Response Function for fMRI

• We propose joint maximum likelihood estimation of HRF and activation levels for fMRI.
• No prior distribution or function for HRF is assumed.
• The proposed solution captures HRF variability across brain regions or subjects.
• The method is developed in the presence of either white noise or colored noise.
• Experiments with synthetic data and real fMRI data demonstrate the merits of the proposed method.

Blood Oxygen Level Dependent (BOLD) functional magnetic resonance imaging (fMRI) maps the brain activity by measuring blood oxygenation level, which is related to brain activity via a temporal impulse response function known as the Hemodynamic Response Function (HRF). The HRF varies from subject to subject and within areas of the brain, therefore a knowledge of HRF is necessary for accurately computing voxel activations. Conversely a knowledge of active voxels is highly beneficial for estimating the HRF. This work presents a joint maximum likelihood estimation of HRF and activation based on low-rank matrix approximations operating on regions of interest (ROI). Since each ROI has limited data, a smoothing constraint on the HRF is employed via Tikhonov regularization. The method is analyzed under both white noise and colored noise. Experiments with synthetic data show that accurate estimation of the HRF is possible with this method without prior assumptions on the exact shape of the HRF. Further experiments involving real fMRI experiments with auditory stimuli are used to validate the proposed method.

Figure optionsDownload high-quality image (110 K)Download as PowerPoint slide