文章摘要

文章信息/Info

Title:: Parameter Sensitivity and Optimization for Hemodynamic ?Response Model of fMRI

Author(s):: YANG Zengyu; JIANG Wentao; CHEN Yu; Department of Applied Mechanics, Sichuan University, Chengdu 610065，China

Keywords:: Functional magnetic resonance imaging; Hemodynamic; Response model; Genetic algorithm; Parameter sensitivity

摘要:: 功能磁共振成像（fMRI）是近几年来研究大脑功能的主要手段，要探究fMRI数据与神经活动的关系，研究血液动力学响应函数模型是关键。由Cohen提出的gamma函数形式的血液动力学响应函数，可以描述相应的响应现象，然而由于在实验数据拟合时未考虑参数间的耦合作用，导致得到的函数模型在求解精度和拟合效果上都并非最优。针对该问题，本研究利用参数敏感度分析和遗传算法相结合的方法，对3参数gamma函数形式的血液动力学响应函数进行整体和全局的优化。首先通过拉丁超立方抽样和spearman秩相关分析理论结合的方法，对该模型中的材料参数进行参数敏感度分析，得出参数敏感度相当的合理取值范围。将该取值范围作为初始搜索区间，结合遗传算法理论对各个参数进行全局最优搜索，结果即为考虑各个参数相互耦合变化的优化值。优化后的模型能较好地描述和预测血液动力学响应函数。该结果较Cohen的模型在拟合效果和实验预测精度上有一定的提高。

Abstract:: Functional magnetic resonance imaging （fMRI) is the main means to study the brain function in recent years, the basis of the imaging principle is to record hemodynamics response caused by neural activity, therefore, to explore the relationship between fMRI data and neural activity, the hemodynamic response function model is the key. The hemodynamic response function model put forward by Cohen can describe corresponding response.However, due to the coupling between parameters is not considered in the experimental data fitting, the model obtained is not optimal in accuracy and fitting effect .To solve this problem, the method of combining parameter sensitivity analysis and genetic algorithm was used to obtain the optimization hemodynamic response function of 3 parameter gamma function form. Firstly, the sensitivity of parameters in the model was calculated by Latin hyper-cube sampling method and spearman rank correlation analysis theory. Then the reasonable range in which the parameters have equivalent sensitivities were obtained. Taking the range as the initial search range, the global optimal search was applied by the genetic algorithm. The optimal parameters were the results considering the coupling effect between parameters. The optimized model predicts the hemodynamic response quite well. In addition, comparing with Cohen’s model, the result makes somewhat improvement in accuracy and data fitting effect.