Tong Zhu1, Michelle Gaugh2, Xiaoxu Liu3, Michael Taylor4, Yuen Tso5, Giovanni Schifitto2, Constantin Yiannoutsos6, Bradford Navia7, Jianhui Zhong8
1Biomedical Engineering, University of Rochester, Rochester, NY, USA; 2Neurology, University of Rochester, Rochester, NY, USA; 3Electrical Engineeering, University of Rochester, Rochester, NY, USA; 44University of California San Diego, San Diego, CA, USA; 5Stanford University, Stanford, CA, USA; 6Indiana University, Indianapolis, IN, USA; 7Tufts University, Medford, MA, USA; 8Imaging Sciences, University of Rochester, Rochester, NY, USA
In a typical neuroimaging multicenter DTI study, biases and variations in data due to differences in scanners among sites prevent pooling of data for conventional statistical inferences. This is one of unsolved critical issues we are facing for multi-center DTI studies. In this study, multiple DTI data of a healthy volunteer were acquired at three imaging centers. Precision of DTI measurement of each center was quantified by the bootstrap analysis of measurement uncertainty while the accuracy (bias) of measurement was evaluated by comparing DTI parameters from each site to those from a super data set with all data combined. Our study suggests that, while precision level of DTI data from different sites is not significantly affected by the short-term variations of scanners at a site, the bias of DTI data from each site will vary and reduce the statistical power when data from multiple sites are combined together. In order to facilitate the multiple-center DTI study, a routine calibration process to regularly measure the bias level is necessary.