vim-1: from Minimally preprocessed data to Estimated BOLD response

I implement the Basis-restricted separable model in the paper "Identifying natural images from human brain activity" to convert Minimally preprocessed EPI data to estimated BOLD responses amplitude on the Sub1_Ses1-5 data. Actually, the deconvolution results have low correlation (low than 0.4) to the data in EstimatedResponse.mat for some voxel (about 40% in V1). This really hinder identification performance. 

Would you provide the codes of Basis-restricted model?  This would be most helpful to my resent work.

Best,

Dai.

Hello Dai (and anyone else with the same question),

We will not be providing the code for the basis-restricted separable model.  However, you may find the MATLAB code in GLMdenoise (http://kendrickkay.net/GLMdenoise/) useful, specifically, the GLMestimatemodel.m function.  This function implements a separable model with a FIR basis, and could be relatively easily tweaked to use a different basis (e.g. sines and cosines).  (GLMdenoise itself provides denoising benefits, but that was not used on the 2008 data.)

Hi, Mark

I found that EPI images and T1 structure images were provided in this data set. It is hard to register EPI images to the T1 structure images cause EPI images was not for the whole brain. Sform matrixes of T1 and EPI were provided in .header file. However, Both of these matrixes match T1 and EPI image to original point in voxel coordinate, which means we could not use inv(Sform_T1)*Sform_EPI transform EPI to T1. How could I match EPI images to T1?

Previously Mark Lescroart wrote:

Hello Dai (and anyone else with the same question),

We will not be providing the code for the basis-restricted separable model.  However, you may find the MATLAB code in GLMdenoise (http://kendrickkay.net/GLMdenoise/) useful, specifically, the GLMestimatemodel.m function.  This function implements a separable model with a FIR basis, and could be relatively easily tweaked to use a different basis (e.g. sines and cosines).  (GLMdenoise itself provides denoising benefits, but that was not used on the 2008 data.)

 

Could you provide the transform matrix match the voxel coordinate in EPI volume to the voxel coordinate in T1 volume?

Previously Mark Lescroart wrote:

Hello Dai (and anyone else with the same question),

We will not be providing the code for the basis-restricted separable model.  However, you may find the MATLAB code in GLMdenoise (http://kendrickkay.net/GLMdenoise/) useful, specifically, the GLMestimatemodel.m function.  This function implements a separable model with a FIR basis, and could be relatively easily tweaked to use a different basis (e.g. sines and cosines).  (GLMdenoise itself provides denoising benefits, but that was not used on the 2008 data.)