Estimating the Hemodynamic Response Timing in Functional MRI Using Coherence Analysis: Theory and Validation

• Main Authors: Chi-Hong Wang, 王志弘
• Other Authors: 陳志宏
• Format: Others
• Language: en_US
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/06968291317159384573

碩士 === 國立臺灣大學 === 電機工程學研究所 === 93 === Functional MRI (fMRI) is a noninvasive technique to study brain function. During the recent decade, it is used widely to determine the spatial layout of brain activation associated with the specified cognitive or sensorimotor tasks. Temporal information in fMRI is less concerned in conventional fMRI analysis because of the sluggish nature of the hemodynamic response. However, with proper experimental design and detail analysis, fMRI can provide the temporal sequence of cortical activation across brain regions during tasks, reflecting the sequence of the individual localized neural events. Present methods to obtain the temporal information in fMRI rely on either rapid sampling rate or assume a specified model. However, with rapid sampling the slice coverage would be reduced. With the specified model, there will be model dependent limitations. All these factors will impair the practicability of the method. In our study, we used coherence analysis to analyze the fMRI voxel time series in frequency domain. It can detect activated areas under rhythmic stimuli. The calculated phase lag can used to estimate the hemodynamic delay. With the devised somatosensory experiments with phase lags in the stimulations, we at first showed the ability of coherence analysis to detect the regions with different hemodynamic delays; in contrast, conventional correlation coefficient method based on temporal domain characteristics might fail to detect areas with large hemodynamic delay. We also demonstrated that the phase lags between different stimuli could be retrieved from the phase lags in the fMRI signals at the corresponding brain regions. We later adopted the event-related paradigm to derive the evoked hemodynamic response function. We then plotted the time delay calculated from the phase lag of the signal from the stimulus paradigm on the fitted curve, which was located near the peak of the curve. We correlated the phase lag with fitted HRF and found good correlation between phase lag from coherence analysis and the time-to-peak in the HRF. To test the consistency and stability of the estimation of the phase and time-to-peak upon different sampling rates, we downsampled a long series of functional activation data acquired at higher rate to simulate the effect of lower sampling rate. The results showed that the phase estimates had higher correlation and lower error in comparison to the original data. In summary, coherence analysis is a powerful method that can be used to detect activated brain region under rhythmic stimulation irrespective of the hemodynamic delay. And it can also estimate the HRF delay in fMRI at the same time. The phase correlates closely with the time-to-peak in fitted HRF, but is less susceptible to the effect of lower sampling rate. In the future, we would like to apply coherence analysis on perfusion-based fMRI contrast to derive more accurate estimate of neuronal activation sequence. We also employ coherence analysis on clinical data in order to estimate possible disease-related alternation in BOLD response temporal dynamics.