thresholding of statistical maps in functional neuroimaging using the false discovery rate（阈值的统计地图功能神经影像使用错误发现率）.pdfVIP

NeuroImage 15, 870–878 (2002) doi:10.1006/nimg.2001.1037, available online at on Thresholding of Statistical Maps in Functional Neuroimaging Using the False Discovery Rate1 Christopher R. Genovese,* Nicole A. Lazar,* and Thomas Nichols† *Department of Statistics, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213; and †Department of Biostatistics, University of Michigan, Ann Arbor, Michigan 48109 Received February 23, 2001 When one uses theoretically motivated thresholds Finding objective and effective thresholds for voxel- for the individual tests, ignoring the fact that many wise statistics derived from neuroimaging data has tests are being performed, the probability that there been a long-standing problem. With at least one test will be false positives (voxels declared active when they performed for every voxel in an image, some correc- are really inactive) among all the tests becomes very tion of the thresholds is needed to control the error high. For example, for a one-sided t test with a 0.05 rates, but standard procedures for multiple hypothe- significance level, the threshold would be 1.645, which sis testing (e.g., Bonferroni) tend to not be sensitive would lead to approximately 1433 voxels declared ac- enough to be useful in this context. This paper intro- tive on average of the 28672 voxels in a 64 64 7 duces to the neuroscience literature statistical proce- image when there is no real activity. The 5% error rate dures for controlli