Predicting the brain response to treatment using a Bayesian hierarchical model with application to a study of schizophrenia
Predicting the brain response to treatment using a Bayesian hierarchical model with application to a study of schizophrenia
复制标题
DOI：
10.1002/hbm.20450
复制
发表时间：
2008-09-01
影响因子：
4.8
通讯作者：
Kilts, Clinton
中科院分区：
文献类型：
--
作者：
Guo, Ying;Bowman, F. DuBois;Kilts, Clinton
体内功能性神经成像，包括功能性磁共振成像（fMRI）和正电子发射断层扫描（PET），在定义精神疾病，如精神分裂症，抑郁症和阿尔茨海默氏病的病理生理学方面变得越来越重要。此外，最近的研究已经开始调查使用功能性神经影像学来指导个体患者治疗选择的可能性。通过研究患者治疗前和治疗后大脑活动之间的变化，研究人员正在深入了解治疗对与特定精神疾病相关的行为相关神经处理特征的影响。此外，这些研究可能揭示了对特定治疗的反应和无反应的神经基础。这些研究的实际局限性在于，治疗后扫描对临床环境中的治疗选择几乎没有指导作用，因为治疗决定先于治疗后脑部扫描的可用性。这一缺陷代表了发展统计方法的动力，该方法将为临床医生提供关于治疗对大脑活动以及最终与神经元相关行为的影响的预测信息。我们提出了一种预测算法，该算法使用患者的预处理扫描，再加上相关的患者特征，预测患者的大脑活动后，指定的治疗方案。我们从贝叶斯分层模型中推导出我们的预测方法，该模型是基于指定训练数据的治疗前和治疗后扫描构建的。我们使用期望最大化算法进行估计。我们使用K折交叉验证评估我们提出的预测方法的准确性，使用我们为神经影像数据提出的两个新措施量化误差。该方法适用于PET和fMRI研究。我们说明其使用PET研究精神分裂症患者的工作记忆和功能磁共振成像数据的例子也提供。
In vivo functional neuroimaging, including functional magnetic resonance imaging (fMRI) and positron emission tomography (PET), is becoming increasingly important in defining the pathophysiology of psychiatric disorders such as schizophrenia, major depression, and Alzheimer's disease. Furthermore, recent studies have begun to investigate the possibility of using functional neuroimaging to guide treatment selection for individual patients. By studying the changes between a patient's pre- and post-treatment brain activity, investigators are gaining insights into the impact of treatment on behavior-related neural processing traits associated with particular psychiatric disorders. Furthermore, these studies may shed light on the neural basis of response and nonresponse to specific treatments. The practical limitation of such studies is that the post-treatment scans offer little guidance to treatment selection in clinical settings, since treatment decisions precede the availability of post-treatment brain scans. This shortcoming represents the impetus for developing statistical methodology that would provide clinicians with predictive information concerning the effect of treatment on brain activity and, ultimately, symptom-related behaviors. We present a prediction algorithm that uses a patient's pretreatment scans, coupled with relevant patient characteristics, to forecast the patient's brain activity following a specified treatment regimen. We derive our predictive method from a Bayesian hierarchical model constructed on the pre- and post-treatment scans of designated training data. We perform estimation using the expectation-maximization algorithm. We evaluate the accuracy of our proposed prediction method using K-fold cross-validation, quantifying the error using two new measures that we propose for neuroimaging data. The proposed method is applicable to both PET and fMRI studies. We illustrate its use with a PET study of working memory in patients with schizophrenia and an fMRI data example is also provided.