Create a connectometry database


This documentation introduces the steps to create a connectometry database. A connectometry database contains information about the subjects' diffusion profile that allows further connectometry analysis.

Step C1: Reconstruct SRC Files for Connectometry


Before this step, please generate SRC files for your study subjects (see Batch Processing using GUI) and place all SRC files in a folder. You will need to run a quality check on all SRC files to make sure that the quality of okay.

Click [Step C1: Reconstruct SRC Files for Connectometry] and select the folder to reconstruct all SRC files. DSI Studio will run QSDR for all SRC files using the HCP-1021 young adult template. If you would like to use study-specific anisotropy template instead of the built-in HCP1021 template, then you will need to reconstruct each SRC file by QSDR in [Step T2: Reconstruction] and assign a different template . Make sure to check [ODFs] in the output option.

Step C1a: Post-reconstruction quality check


Each FIB file generated from the previous step will have a file name such as *.odf8.f5rec.cdm.qsdr.1.25.2mm.R67.fib.gz

The R67 indicates that the R-squared value between the subject and the template's QA map is 0.67. 

**An R-squared value lower than 0.6 may require further inspection to confirm whether it is due to registration error**.

The most common cause for a low R2 is an inverse order of the axial slices, which can be corrected in the reconstruction step to flip the image volume at the Z direction. 
The second common cause is a prominent artifact in the background, which may be handled by introducing a brain mask and use [Edit][Trim Image] in the reconstruction step to clean up the background. 

If you still cannot solve the low R2 value problem, please feel free to send the data to me (uploaded provided in the webpage of the Discussion forum), and I will figure out a solution for you.

Step C2: Create a connectometry database


A connectometry database requires the ODF-containing fib files for all subjects and an atlas that serves as the sampling skeleton.

1) In the [Diffusion MRI Connectometry] tab, click "Create a connectometry database".

This brings up the following dialog.


2) Click the [Search in Directory] and select the study folder. You may also use the "Add" button to select individual FIB file. You can change the order of the files by the "up", "down", "sort" button. You can save/load a list of the SRC files.

*If you are going to study the change in a longitudinal study, make sure that you place baseline and followup scans of the same subject together (e.g. baseline of subject#1, follow-up of subject#1, baseline of subject#2, followup of subject#2...etc.)

3) Assign the templatefile (skip if a default file is assigned). The default atlas is HCP1021 2-mm atlas provided in the DSI Studio package, which is available at Atlas and sample images

You can also create your own atlas, and assign it here so that DSI Studio will use it to sample anisotropy values. 

*If your DWI were acquired at a high resolution (e.g. 1.4mm), QSDR will reconstruct data to 1-mm MNI space, and you will need to use HCP1021 1-mm template (available at Atlas and sample images).
*If your data have both 2-mm MNI FIB file and 1-mm MNI FIB, DSI Studio will prompt an "inconsistency resolution" error. To solve it, downsample the high-resolution SRC files in STEP T2 Reconstruction using [Edit][Resample to isotropic] (assign 2 mm), and rerun QSDR reconstruction.

4) Assign "Index of interest": The default setting is "QA" as known as the quantitative anisotropy. You can also study other diffusion measures such as FA, AD, RD, MD, RDI, NRDI (see here for explanation for each index) using connectometry.

*The QA here is the SDF values sampled at fiber orientations, and the fiber orientations here are defined by the template in the previous steps (please refer to the local connectome fingerprint paper at https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1005203). The sampled values are similar to QA defined in the native space but there are differences. QA in the native space is defined on the peaks of SDF, but here the QA values are extracted at the template fiber orientationsThe physical meaning is the same, but their evaluation approach is different.


4) Specify the output file name.

    Please assign the file extension as *.db.fib.gz

5) Click the "Create Database" button to create the connectometry database as a db.fib.gz file. You can add or remove subjects from the database using [Diffusion MRI Connectometry][Edit Connectometry Database]


At the end of this step, you will have a connectometry database, which is a file name with extension *.db.fib.gz

TIP: You can append more subjects or remove subjects from a database later using [Tools][O1:Open a Connecometry Database] from the main window tab.

TIP: If you are going to study the change in a longitudinal study, use [Tools][O1:Open a Connecometry Database] to calculate the difference between baseline scan and followup scan for each subject (e.g. base of s#1, followup of #1, base of s#2, followup of s#2...etc.) and save a new database.

Optional: Create a population average template

The FIB files created from Step C1 can be averaged into a population average template using the following steps:

1) Click [Tools][O8: Create template/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 template for creating a connectometry database.

Reference

[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: 


Optional: Output local connectome fingerprint


What is local connectome fingerprint?


DSI Studio provides a way to sample white matter characteristics as a "fingerprint", an approach called "local connectome fingerprinting" (LCF) [1]. The method uses on q-space diffeomorphic reconstruction (see Reconstruction(DTI, QBI, DSI, GQI, QSDR) to calculate the density of diffusing water along major fiber bundles from diffusion MRI. LCF provides a vector measurement characterizing white matter structures at the voxel level for each dMRI scan. LCP was found to be highly unique for each individual [1], making it a tangible metric of connectomic differences. The variability in LCF was also found to be partially determined by genetic factors, but largely plastic across time. LCF potentially opens a new door for probing the influence of pathological, genetic, social, or environmental factors on the unique configuration of the human connectome. The density sampled exhibits high individuality that allows us reliably identify individuals or quantify similarity between two connectome architectures.


Output local connectome fingerprint


Click [Tools][O1: Open a Connectometry Database] to bring up the connectometry database content. Then click on [Fingerprinting][Save fingerprints] to export the fingerprints of the whole brain. There are several options:

  • If you would like to output the fingerprint of a specific region, load a ROI NIFTI file using [Fingerprinting][Load fingerprint mask...] to assign an ROI as a mask, and then save the fingerprint using [Fingerprinting][Save fingerprints]. The exported fingerprint will use the mask to extract the values.
  • Fiber threshold is a value used to filter out gray matter region. The default is 0.6 otsu's threshold.
  • Normalize FP will scale the fingerprint values so that the standard deviation of a fingerprint vector is scaled to one.
  • The fingerprints will be stored in a MATLAB mat file with the following structures.

dimension: the dimension of the template space.
'subject0' to 'subjectN': a total of N subjects' fingerprint vectors
subject_names: the name for each subject. It is a 1xN vector of number and can be converted to named by using the command, ID = textscan(char(subject_names),'%s');
voxel_location: an 1-by-N matrix storing an index for each fingerprint entry.
mni_location: a 3-by-N matrix the MNI coordinate for each fingerprint entry
fiber_direction: a 3-by-N matrix storing the associated fiber direction for each fingerprint entry.

Reference

[1] Yeh F-C, Vettel JM, Singh A, Poczos B, Grafton ST, Erickson KI, et al. (2016) Quantifying Differences and Similarities in Whole-Brain White Matter Architecture Using Local Connectome Fingerprints. PLoS Comput Biol 12(11): e1005203. doi:10.1371/journal.pcbi.1005203 (link)



Comments