Documentation‎ > ‎

Construction of a high angular resolution atlas


DSI Studio is able to reconstruct diffusion properties in a standard space using q-space diffeomorphic reconstruction (QSDR) and average them to create a group-average diffusion template for tracking and analyzing representative fiber pathways. 

The following are steps to create a diffusion template using DSI Studio.

Convert diffusion data to src files

You may use either the DICOM or NIFTI files to create SRC files. An SRC file stores diffusion images and the gradient table. You can also use command line functions of DSI Studio to batch converting all subject data to src files.

Q-space diffeomorphic reconstruction (QSDR)

Click on "STEP2: Reconstruction" and select ALL .src files. In the reconstruction windows, jump to "select reconstruction method". QSDR uses the mask from the build-in template, and you can ignore the mask assignment. 

Select QSDR as the reconstruction method. Here are some parameters that may affect the results.

Diffusion sampling length ratio

this ratio defines the sensitivity of the reconstruction method to fast diffusion. A higher value gives better crossing fiber resolving, but it also increases the sensitivity to noise. A value around 1.2 gives good results, but the optimal value depends on the b-table used here. I would recommend using GQI to test out the best length ratio first and apply the optimal setting here for QSDR.

Registration method

SPM norm 7-9-7 is the entry level nonlinear mapping method. The computation is fast (2~3 minutes), but its accuracy is limited. A higher order of normalization gives better results. I recommend using the highest order possible for template creation. DSI Studio can make use of multi-core CPU to speed up the registration, and you may choose the number of thread at the bottom of the window.

Output resolution

This assigns the resolution of the atlas. A 2-mm or 1-mm output resolution can be selected here.

ODF sharpening

The ODF sharpening is in the advanced options. QSDR can be combined with diffusion deconvolution [2] or diffusion decomposition [3] to construct a fiber ODF atlas. Diffusion deconvolution using L2 regularization to compute the fiber ODF. The sharpening result is smoother. Diffusion decomposition using L1 regularization and assume sparsity in fiber ODF. It gives a pointy profile of fiber ODF. The response function can be selected automatically. This is the default and recommended setting. Alternatively, one may assign the coordinate of the voxel for sampling the response function.


You must check "ODFs" in the output box. This will output full ODF data in the fib file that allows further averaging. You may choose to output spatial mapping function and the jacobian determinant. They are however not necessary for constructing an atlas.

Parallel reconstruction

DSI Studio can make use of multi-thread to speed up reconstruction. A higher value is recommended. Once you confirm the setting, click "Run reconstruction". DSI Studio will reconstruct ALL src files you selected. This will generate a fib file for each src file.

Averaging ODFs

1) In the connectometry toolbox, click "Step 1: create skeleton". 

2) Add in all FIB files and specify the output file name. This dialog outputs two fib files, one without and one with the averaged ODFs. The one without the ODFs can be used as the local tract skeleton for creating a connectometry database.

[1] Yeh, F.C., Tseng, W.Y., 2011. NTU-90: A high angular resolution brain atlas constructed by q-space diffeomorphic reconstruction. Neuroimage 58, 91-99. (pdf)
[2] Yeh, F.C., Wedeen, V.J., Tseng, W.Y., 2011. Estimation of fiber orientation and spin density distribution by diffusion deconvolution. Neuroimage 55, 1054-1062. (pdf)
[3] Yeh, F.C., Tseng, W.Y.Sparse Solution of Fiber Orientation Distribution Function by Diffusion Decomposition”, PLoS One. 2013 Oct 11;8(10):e75747. doi: 10.1371/journal.pone.0075747. (pdf).