Geissler Alexander, Gartus Andreas, Foki Thomas, Tahamtan Amir Reza, Beisteiner Roland, Barth Markus
Clinical fMRI Study Group, Departments of Neurology, Neurosurgery, and Radiology, Medical University of Vienna, Waehringer Guertel 18-20, 1090 Vienna, Austria.
J Magn Reson Imaging. 2007 Jun;25(6):1263-70. doi: 10.1002/jmri.20935.
To evaluate the impact of data quality on the localization of brain activation in functional magnetic resonance imaging (fMRI) and to explore whether the temporal contrast-to-noise-ratio (CNR) provides a quantitative parameter to estimate fMRI quality.
We investigated two methods for defining the CNR by comparing them on a single-run, single session, as well as on a group-wise basis. The CNRs of healthy subjects and a group of patients with brain lesions were calculated using two different strategies: one based on a general linear model (GLM) analysis (CNR_SPM), and one that acts as an adaptive low-pass filter and assumes that the high-frequency components contain the temporal noise (CNR_SG). Runs with low CNR were identified as outliers using a common exclusion criterion (2 x standard deviation (SD)).
The results of the two CNR methods are highly correlated. Both between and within subjects and patients the CNR showed quite large variations, but the average CNR did not differ between a group of healthy subjects and a patient group. In total, seven of 213 runs (3.3% of all runs) had to be excluded when CNR_SG was used, and 14 of 213 (6.6%) runs had to be excluded when CNR_SPM was used.
Calculating the CNR using an adaptive low-pass filter gives similar results to a GLM-based approach and could be advantageous for cases in which the hemodynamic response function (HRF) differs significantly from common assumptions. The CNR can be used to identify and exclude runs with suboptimal CNR, and to identify sessions with insufficient data quality. The CNR may serve as a quantitative and intuitive parameter to assess the performance and quality of clinical fMRI investigations, including information on both functional performance (contrast) and data quality (noise caused by the system and physiology).
评估数据质量对功能磁共振成像（fMRI）中脑激活定位的影响，并探讨时间对比噪声比（CNR）是否提供了一个定量参数来估计fMRI质量。
我们通过在单次运行、单一会话以及组水平上比较两种定义CNR的方法进行了研究。使用两种不同策略计算健康受试者和一组脑损伤患者的CNR：一种基于一般线性模型（GLM）分析（CNR_SPM），另一种作为自适应低通滤波器，假设高频成分包含时间噪声（CNR_SG）。使用通用排除标准（2倍标准差（SD））将CNR低的运行识别为异常值。
两种CNR方法的结果高度相关。无论是在受试者之间还是患者内部，CNR都表现出相当大的变化，但一组健康受试者和患者组之间的平均CNR没有差异。使用CNR_SG时，213次运行中有7次（占所有运行的3.3%）必须排除，使用CNR_SPM时，213次运行中有14次（占6.6%）必须排除。
使用自适应低通滤波器计算CNR与基于GLM的方法得到的结果相似，对于血流动力学响应函数（HRF）与常见假设显著不同的情况可能具有优势。CNR可用于识别和排除CNR次优的运行，并识别数据质量不足的会话。CNR可作为一个定量且直观的参数来评估临床fMRI研究的性能和质量，包括功能性能（对比度）和数据质量（由系统和生理引起的噪声）方面的信息。
