请教各位老师关于审稿人对于灰质体积使用Alphasim矫正的问题,谢谢

我在对灰质体积结果进行多重比较矫正的时候使用了3dClustSim,审稿人对这种方法提出了下面的问题,不知道应该如何回答审稿人的意见为好,谢谢。

I think the corrections for multiple comparison based on cluster size in VBM should be performed with permutation based methods or non-isotropic adjusted cluster size [for related discussion, see (Hayasaka et al., 2004)]. But I am not sure Monte Carlo simulation is basically a permutation based method or not. Please clarify. In case, non isotropic adjusted cluster size of SPM5 is used, authors have to take care of the conditions revealed by Silver et al. (2011, Neuroimage) (Though, I am not sure about conditions for non isotropic adjusted cluster size of SPM8).

Monte Carlo simulation is a non-parametric non-isotropic method for determining FWE threshold, and has the same mathmetical mechanism with permutation (there are some minor differences among permutation, bootstrap and Monte Carlo simulation methods).

I don't know what's the issue in  Silver et al. (2011, Neuroimage). Please verify. 

应该是这篇文章中提到的问题,我也看不大明白,审稿人好像对Alphasim没有好感,我的第一次的回复为:

Point 1: First, I think the corrections for multiple comparison based on cluster size in VBM should be performed with permutation based methods or non-isotropic adjusted cluster size [for related discussion, see (Hayasaka et al., 2004)]. But I am not sure Monte Carlo simulation is basically a permutation based method or not. Please clarify. In case, non isotropic adjusted cluster size of SPM5 is used, authors have to take care of the conditions revealed by Silver et al. (2011, Neuroimage) (Though, I am not sure about conditions for non isotropic adjusted cluster size of SPM8).
Response:
Thanks for your suggestion. In this study, we used AlphaSim program for multiple comparison corrections. AlphaSim program is part of a standard neuroimaging toolbox, AFNI (http://afni.nimh.nih.gov/) and is one of the methods for multiple comparison correction combining voxel intensity and cluster extent. In AlphaSim, Monte Carlo
permutation(Ward, 2000). simulations are used to estimate the null distribution. Specifically, in the program, the algorithm generates an estimate of the overall significant level achieved for various combinations of probability threshold and cluster size threshold by iteration of the process of random image generation, Gaussian filtering, thresholding, image masking, and tabulation of cluster size frequencies

Although the algorithm of AlphaSim is not exactly the same with the algorithm mentioned in Hayasaka et al. (2004), we think, it does provide the appropriate cluster-level threshold to achieve the desired false-positive rate. Moreover, AlphaSim is widely used in published literatures about VBM (DeYoung et al., 2010; Ding et al., 2012; Farb et al.; Kong et al., 2013; Schwartz et al., 2010; Yang et al., 2013; Zou et al., 2012) results.

编辑给我们的回复为:

For the comment 1, I understood the situation, if the simulation was performed with the randomly generated images, then the results may have the similar problems that have been reported in Silver et al., (2011, Neuroimage)'s study. In other words, it is shown for whatever the reasons, randomly generated images (at least those created by Hayasaka et al) give different results when compared with the results of real brain structural images. But that is just the possibility and unlike original Hayasaka et al., (2004)'s cluster size test for VBM, the present methods have not been proven to be false and it is established and widely used and I am not entitled to deny the results with such methods without scientific data. So, I recommend authors to cite the abovementioned points as the possible limitation of this study. In my opinion without scientific basis, authors may better rely on other multiple comparison correction methods in the future studies before someone prove the monte calro simulation's application to VBM is wrong with scientific data, I have never used this method and don't know the details about this method but often "feel like" the other researchers' results given by this monte calro simulation are lenient and woder why.