Documentation‎ > ‎

How to acquire DSI and multishell data


Multishell acquisition can be done by acquiring two or more DTI with different b-value. Here is a working example on a SIEMENS 3T scanner: 

One DTI dataset has 64 directions. b-value = 3000 sec/mm2, and TR/TE = 6300 ms/121 ms (as short as possible). 
Another DTI dataset has 30 gradient directions. b-value = 1500 sec/mm2, and TR/TE = 5500 ms/101 ms (as short as possible). 

DSI Studio will correct for different T2-shine through due to different TE.

How to acquire DSI

Diffusion spectrum imaging requires a specific grid-like b-table structure. I recommend acquiring "DSI-101" with a maximum b-value of 4,000 s/mm2. An example of the b-table is provided at the bottom of this page (file name: dsi101_bmax4000_half.txt). If you are using a SIEMENS scanner, there is a vector table file provided ("DSI101_B4000_DiffusionVectors.txt"). To use the table, you need to use the "free" mode on the Diff tab and input 4,000 in the "b-value 2" field. Use the shortest possible TR and TE to acquire diffusion images.

Acquiring DSI on a Bruker scanner is more challenging because each DWI section can only acquire one b-value for a set of directions. You may need to combine multiple sections together as one DSI. To do this, first take a look at the DSI 101 b-table, dsi101_bmax4000_half.txt, provided at the bottom of this page. The first section will acquire a b-value of 307.692 at (-1, 0, 0),(0 1 0),(0 0 1). Then second section will acquire a b-value of 615.385 at (-0.707107,0.707107,0), (-0.707107,0,0.707107)...etc. All section should use the same receiver gain and the same TE (shortest possible). You may use a spin echo sequence or a stimulated echo sequence. The diffusion separation can range between 10 ms to 80 ms.


You may also create a customized DSI table for SIEMENS scanner using gtable.zip provided at the bottom of this page. Alternatively, you may follow the following steps to design a new DSI b-table. 

(1) Determine the max_q, e.g. |q|^2 < 13. A higher max_q value results in more sampling directions. Here the q is an integer representing the location of the Fourier transform matrix.
(2) iterate all possible q vectors from the origin. e.g., (0,0,0), (0,0,1), (0,1,0), (1,0,0), .... 
(3) Determine the maximum b-value. e.g. 4000 s/mm2
(4) calculate the b-value for each q vector by scaling b-value with respect to |q|^2.

Here is the background information for building a DSI b-table. 

1) The Fourier transform in DSI uses only diffusion sensitization gradient (termed q), not the b-value. Thus we need to use q to design the diffusion scheme and convert q vector table to b-table. The q vector should be arranged at Cartesian points (thus termed grid scheme) to facilitate Fourier transform. An example of the q table is attached at the bottom of this webpage (dsi_q_vector_203.txt and dsi_q_vector_515.txt). One should note that b-value scaled with q^2, not |q|. b-table also scales with diffusion time. It is a product of both diffusion gradient strength and diffusion time.

2) The Fourier relation in DSI assumes long diffusion time (big delta) and short diffusion encoding duration (small delta). In practice, the optimal setting is using stimulated echo to get a long diffusion time while keeping TE small. However, most studies still use spin echo with diffusion time close to diffusion encoding duration. The resulting PDF may not have a correct length scale, but the peak ODF directions are not affected by this setting. 

3) q has a unit in micron^-1. However, since q vectors are placed at Cartesian points in space, they are usually represented by integers. e.g. (1,0,0), (0,1,0),...etc. The actual diffusion gradient strength and the q-coding differs only by a scalar factor. In DSI reconstruction, only the q interger is used to place diffusion MRI signals in a FFT matrix. The actual q values or b values are not used in DSI reconstruction. (see DSI reconstruction matlab code here).

4) The signal at q is assumed to be equal to -q because the diffusion PDF (termed averaged propagator in q-space imaging) is assumed to be symmetric with respect to the origin. Thus a sampling scheme can acquires only a half sphere of the q vectors. Those schemes are called half-sphere schemes, as opposed to full schemes that acquire all q vectors in q-space (which is redundant)

5) The signals at high q has very low SNR and a DSI scheme is often designed by acquiring all q vectors that satisfies |q|^2 < max_q. Here the q vectors use integer form and max_q can be arbitrary. DSI reconstruction will pad high b-values entries (greater than max_q) with zeros. 

Additional suggestion


1. If you want to do fiber tracking, acquire isotropic voxel. 
2. Use shortest possible TE/TR



SelectionFile type iconFile nameDescriptionSizeRevisionTimeUser
ċ

Download
  2k v. 1 Aug 3, 2012, 10:38 AM Fang-Cheng Yeh
ċ

Download
  4k v. 1 Aug 3, 2012, 10:38 AM Fang-Cheng Yeh
SelectionFile type iconFile nameDescriptionSizeRevisionTimeUser
ċ

Download
  3k v. 2 Jun 15, 2012, 12:50 AM Fang-Cheng Yeh
ċ

Download
  11k v. 2 Jun 15, 2012, 12:50 AM Fang-Cheng Yeh
ċ

Download
  6k v. 2 Sep 22, 2013, 8:24 PM Fang-Cheng Yeh
ċ

Download
  14k v. 2 Jun 15, 2012, 12:50 AM Fang-Cheng Yeh
ċ

Download
  16k v. 2 Jun 15, 2012, 12:50 AM Fang-Cheng Yeh
SelectionFile type iconFile nameDescriptionSizeRevisionTimeUser
ċ

Download
This is the b-table you can put into the SIEMENS pulse sequence  6k v. 2 Dec 14, 2015, 6:12 PM Fang-Cheng Yeh
ċ

Download
  8k v. 2 Mar 23, 2015, 2:19 PM Fang-Cheng Yeh
ċ

Download
Run this program to generate SIEMENS Free Model vector table  330k v. 3 Mar 23, 2015, 2:18 PM Fang-Cheng Yeh
Comments