Marrelec Guillaume, Ciuciu Philippe, Pélégrini-Issac Mélanie, Benali Habib
INSERM U494.
Inf Process Med Imaging. 2003 Jul;18:635-46. doi: 10.1007/978-3-540-45087-0_53.
A convenient way to analyze BOLD fMRI data consists of modeling the whole brain as a stationary, linear system characterized by its transfer function: the Hemodynamic Response Function (HRF). HRF estimation, though of the greatest interest, is still under investigation, for the problem is ill-conditioned. In this paper, we recall the most general Bayesian model for HRF estimation and show how it can beneficially be translated in terms of graphical models, leading to (i) a clear and efficient representation of all structural and functional relationships entailed by the model, and (ii) a straightforward numerical scheme to approximate the joint posterior distribution, allowing for estimation of the HRF, as well as all other model parameters. We finally apply this novel technique on both simulations and real data.
一种分析血氧水平依赖性功能磁共振成像（BOLD fMRI）数据的便捷方法是，将整个大脑建模为一个以其传递函数——血流动力学响应函数（HRF）为特征的平稳线性系统。尽管HRF估计是最受关注的，但由于该问题是病态的，目前仍在研究中。在本文中，我们回顾了用于HRF估计的最通用贝叶斯模型，并展示了如何将其有益地转化为图形模型，从而实现：（i）清晰有效地表示模型所涉及的所有结构和功能关系；（ii）一种直接的数值方案来近似联合后验分布，从而能够估计HRF以及所有其他模型参数。我们最终将这种新技术应用于模拟数据和实际数据。
