INTER-SITE QA DEVELOPMENT FOR THE FIRST BIRN

本子项目是利用由NIH/NCRR资助的中心赠款提供的资源的众多研究子项目之一。子项目和研究者（PI）可能已经从另一个NIH来源获得了主要资金，因此可以在其他CRISP条目中表示。列出的机构是中心的，不一定是研究者的机构。FIRST（功能成像研究精神分裂症试验台）生物医学信息学研究网络（fBIRN）项目是有史以来第一个大规模、多中心的精神分裂症功能磁共振研究（http://www.nbirn.net）。fBIRN项目将汇集来自11个参与地点的功能磁共振成像数据，以便在短时间内获得大量不同的研究人群。为了保持扫描仪的稳定性，本中心制定了在所有站点使用的定期质量保证（QA）方法。在每个站点定期进行扫描，并将数据上传到网络服务器。编写了ISMRM 2004的摘要并被接受。在纵向功能磁共振成像研究中，定期监测扫描仪性能以确保稳定性、几何精度、信噪比（SNR）等特性保持准确和一致是非常重要的。当来自多个扫描器的数据在初始和较长时间内进行比较或合并时，例如在fBIRN项目中，稳定性更为重要。本中心开发的质量保证方法适用于fBIRN。其中最重要的是检查扫描仪的稳定性，但也得出信噪比，并在这里描述。每周稳定性测试使用fBIRN fMRI扫描方案，直径为17 cm的琼脂幻影（35轴，4 mm连续切片，22 cm FOV， 64x64矩阵，TE 30ms/40ms (3T-4T/1.5T), TR 3000ms， 200个时间帧，10分钟扫描时间，EPI或螺旋）。琼脂糖凝胶幻影用NaCl掺杂以呈现典型的头部RF负载，并且优选于水以避免旋转伪影。仅使用来自中心切片（18）的时间序列数据进行分析。使用以图像为中心的20x20 ROI来测量序列中每个时间帧的图像强度。二次去趋势后计算均方根分数波动，从趋势线极值处得到漂移（图1，顶部）。计算傅里叶谱（图1，中间）。使用大小从1x1到20x20不等的roi进行“Weisskoff分析”，如图1所示。在此分析中，标准偏差（STD）应减小为？（ROI中的体素数），如果噪声不相关。来自RF或梯度源的系统不稳定性往往会引起低空间频率的图像波动，这种波动表现为体素之间的相互关联。这反过来又导致STD偏离对1/（ROI宽度）的线性依赖，并随着ROI大小的增大而触底，如图1所示（参见虚线）。请注意，顶部的图是用最大的ROI大小执行的，有意选择它来表示系统有限的（而不是SNR有限的）噪声。此外，信号与波动噪声比（SFNR）由（时间序列均值图像）/（时间序列标准差图像）构成的图中的ROI测量。最后，在噪声图像中减去偶数时间帧的平均值和奇数时间帧的平均值，并除以平均信号，从ROI测量的噪声中计算信噪比。

This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. The subproject and investigator (PI) may have received primary funding from another NIH source, and thus could be represented in other CRISP entries. The institution listed is for the Center, which is not necessarily the institution for the investigator. The FIRST (Functional Imaging Research Schizophrenia Testbed) Biomedical Informatics Research Network (fBIRN) program is the first-ever large scale, multi-center fMRI study of schizophrenia (http://www.nbirn.net). The fBIRN project will pool fMRI data from each of the 11 participating sites to enable the acquisition of a large and diverse study population in a modest time period. To maintain scanner stability, our Center developed methods in use at all the sites for periodic quality assurance (QA). Scans are performed regularly at each of the sites and data uploaded to the network servers. An abstract was prepared and accepted for ISMRM 2004. For longitudinal fMRI studies, it is important to periodically monitor scanner performance to assure that stability, geometric accuracy, signal to noise ratio (SNR) and other characteristics remain accurate and consistent. Constancy is even more important when data from multiple scanners are being compared or combined both initially and over extended periods of time, such as in the fBIRN project. QA methods developed at our Center were adapted for fBIRN. The most important of these examines scanner stability but also derives SNR and is described here. Data Acquisition The weekly stability test uses the fBIRN fMRI scan protocol with a 17 cm diameter agar phantom (35 axial, 4 mm contiguous slices, 22 cm FOV, 64x64 matrix, TE 30ms/40ms (3T-4T/1.5T), TR 3000ms, 200 time frames, 10 minute scan time, EPI or spiral). The agarose gel phantom was doped with NaCl to present RF loading typical of a head and was preferred over water to avoid swirling artifacts. Analysis Only timeseries data from the central slice (18) is used for analysis. Image intensities are measured for each time frame in the series using a 20x20 ROI centered in the image. The RMS fractional fluctuation is calculated after quadratic detrending, and drift is obtained from extrema of the trend line (Fig. 1, top). The Fourier spectrum is calculated (Fig. 1, middle). A 'Weisskoff analysis' is performed using ROIs with sizes varying from 1x1 to 20x20, as shown in Fig. 1, bottom. In this analysis, the standard deviation (STD) should decrease as ?(number of voxels in ROI) if the noise is uncorrelated. System instabilities from RF or gradient sources tend to cause low spatial-frequency image fluctuations, which are manifested as cross-correlation between voxels. This in turn causes STD to depart from linear dependence on 1/(ROI width) and hit a floor as the ROI size becomes larger, as shown in Fig. 1 (cf. with dashed line). Note that the top plot is performed with the largest ROI size, intentionally picked to represent the system limited (rather than SNR limited) noise. In addition, the signal to fluctuation noise ratio (SFNR) is measured by an ROI in a map made from (timeseries mean image)/(timeseries standard deviation image). Finally, the SNR is calculated from noise measured with an ROI in the noise image obtained by subtracting the average of the even time frames and average of the odd time frames and dividing into the mean signal.