Diffusion MRI Fiber Tracking in DSI Studio


DSI Studio provides a deterministic fiber tracking algorithm that uses quantitative anisotropy to improve accuracy [4]. Once the tractography is generated, the results can be further analyzed using track-specific analysis or network measures.

Started from the new version from April 2020, DSI Studio provides automatic fiber tracking for human studies to greatly simplify the manual fiber tracking process. A quick tutorial video is shown on the right. To process multiple subjects, there is also [Batch Processing][B4 Automatic Fiber Tracking] (see here for details). 

For those choose to use manual ROI-based fiber tracking, details protocol for different pathways can be found in the TrackEM project (https://my.vanderbilt.edu/tractem/protocol/https://www.biorxiv.org/content/10.1101/651935v2.full) and recent TRACULA update (https://www.biorxiv.org/content/10.1101/2021.06.28.450265v2.full)

Step T3: Fiber tracking

To run "manual" fiber tracking in DSI Studio, open the main window and click a button named [Step T3: Fiber Tracking] to select a FIB file. DSI Studio will bring up the tracking window. 

The standard steps to create a tractography are as follows
Step T3a and T3b: Assign regions as seeding areas, region of interest, region of avoidance, or ending regions.
Step T3c: Adjust tracking options
Step T3d: Run fiber tracking
The information for each process is detailed in the following sections.

Step T3a: Assign Regions and Step T2b: Draw Regions

The upper left tab shows a region list under [Step T3a: Assign Regions]. The regions can be used in fiber tracking to place seeds or limit tracking results.
There are several ways to assign regions:

Load from NIFTI files

A NIFTI file can be loaded in the following ways.

For FIB reconstructed by DTI or GQI:

1. If the image size of the NIFTI file matches the diffusion data, then no image resampling is applied. DSI Studio will load it directly.
2. If the ROI file is originated from subject's own T1W or T2W (e.g. a FreeSurfer segmentation in T1-weighted or T2-weighted image space), then you need to first insert the original structural image (e.g. T1-weighted or T2-weighted images) by [Slices][Insert T1/T2]. DSI Studio will work out a background registration, which may take 5 minutes to converge. After the registration is stabilized, load the ROI file, and DSI Studio will automatically apply the registration from the previous structural image. The transformation matrix in the nifti header will not be used here.
3. If the NIFTI file is in the MNI space, use [Region][Load MNI Region] to import the region. A background nonlinear registration will be conducted that allows wrapping the ROI to the subject space.

For FIB reconstructed by QSDR:

1. If the dimension of the NIFTI file matches that of the raw diffusion data, DSI Studio will automatically warp the region to the QSDR space.
2. If the NIFTI file was generated from T1W (e.g. manually drawing or from Freesurfer), the NIFTI file has to be converted to the original diffusion space. Then DSI Studio can load it and apply transformation automatically.

DSI Studio can take the nifti file with multiple values as multiple ROIs. You can supply a label file with the nifti file. An example of the label file can be found under the "atlas" folder in the DSI Studio package.

Load from atlases

DSI Studio provides a list of atlases that can be added to the region list. Users can add anatomical landmarks by clicking on the [Aatlas...] button in [Step T3a]. DSI Studio will perform a nonlinear registration to bring atlas to the subject space.

Add a new atlas

An atlas is an integer-valued image volume that records the location of each brain region.

For human atlas, please copy the MNI space (presumably ICBM2009a nonlinear space) to \atlas\ICBM152 folder. For animal atlas, please find the corresponding folder, such as the one for mouse, rat, marmoset, or rhesus.

In the Mac version, those folders are stored in the app package (Right-click on dsi_studio_64.app to open the DSI Studio package and look for the template and atlas folder). 

After copying a new atlas to the \atlas\ICBM152 (this location is for humans), restart DSI Studio to see the new atlas added to the ICBM152 menu.

Add a new template

A template is a population-average image volume that defines a standard space. The template has its imaging modality, such as T1W, T2W, anisotropy map.

The template used in DSI Studio will be an anisotropy map (e.g. FA map or QA map) and also an isotropy map (ISO map).

DSI Studio can add a new template for different animal species or different subject populations (e.g., neonatal, elder...etc.). 

To add a template, first, create a folder under the /atlas folder (e.g. /atlas/NAME_OF_TEMPLATE) and then copy the anisotropy map to the folder and named it NAME_OF_TEMPLATE.QA.nii.gz 

DSI Studio also needs an "ISO" map to utilize dual-modality normalization. The file should also be placed under /atlas/NAME_OF_TEMPLATE and should be saved as NAME_OF_TEMPLATE.ISO.nii.gz 

An ISO map is optional, but having an ISO template with the QA template will greatly increase the normalization accuracy. 

If you only have a T1W template (e.g. child or elder population), then you can warp ICBM152 QA and ISO template to your template space using the following procedure:

1. [O6: Linear registration box] select "\atlas\ICBM152\ICBM152.QA.nii.gz" as the subject image and "\mni_icbm152_t1_tal_nlin_asym_09c.nii.gz" (provided in DSI Studio package, windows version) as the reference image. 
2. Click on [Save Warped Image] and save it as a new file. Let's name it new.QA.nii.gz
3. Repeat steps 1 and 2 but select "\atlas\ICBM152\ICBM152.ISO.nii.gz" to create new.ISO.nii.gz
4. [07: Nonlinear registration box], [Open Subject]->"\mni_icbm152_t1_tal_nlin_asym_09c.nii.gz" and [Open Reference]->your new t1w template. I would suggest using smoothness=1.0 or higher to avoid over-distorted results. Click [Run]. You may experiment with different smoothness values to get the best result.
5. After registration, select the top menu [File][Apply Warpping]->new.QA.nii.gz and overwrite it.
6. Repeat 5 but select new.ISO.nii.gz and overwrite it.
7. Now you have new QA and ISO templates in your T1W template space. After adding them to \atlas\NAME_OF_TEMPLATE, restart DSI Studio to see the template added to the menu.

Load from text coordinate files

DSI Studio can load ROIs from a text file of the ROI coordinates. The coordinates should be integers (floating point will be rounded up) because the coordinates indicate the "voxel" of the seeding regions. Users should be noted that the actual seeding points are uniformly distributed "within" the seeding region. The coordinates here may not necessarily match the exact seeding points in the tracking algorithm.

Save NIFTI files for FSL tracking

The region in DSI Studio can be saved in different resolution (e.g. ROI drawn in imported T1W). Also, there is a difference in the orientation convention between FSL and DSI Studio. As a result, the nifti file saved by DSI Studio may not be readily used by FSL protrackx. 

To handle this problem, the region needs to be converted to the diffusion space if it was first drawn in an imported image volume:

1. Load the FIB file and start a new region by [Region][New region] before inserting any T1W/T2W.
2. Load the DSI Studio ROI file and merge the region with the new region created in step1 using [Regions][Merge all]

Now the ROI will have the same resolution and dimension as the diffusion volume. The next step is to handle the orientation difference:

3. Flip image by X and Y using [Regions][Move region][Flip X] and [Flip Y]
4. Save the image as an nifti file using [Region][Save region as...]

The exported region now can be used in FSL.

Manually draw a region

YouTube Video

Video: how to "semi-automatically" segment a lesion in the brain using tools provided in DSI Studio.

Tip:  check out shortcuts at the bottom of this page.

You can manually draw a region in the region window to the left. A toolbar on the top showing different drawing tools. To draw a region in the region window, left-click in the window and drag to create a new region. Any further click will add voxels to the existing region. To remove part of a region, right-click to assign the region to be erased. 

The function of each tool is detailed as follows:
1. The rectangle tool draws a rectangular region.
2. The free-handle tool draws a shape using the cursor trajectory.
3. The polygon tool draws a polygon region.
4. The ball bool draws a ball in the 3D space.
5. The cubic tool draws a cubic in the 3D space.
6. The all-way-cross tool allows for dragging a region, or the slices.
7. The ruler can be used to estimate the distance in mm. 

The region can be assigned as "seed","ROI","ROA",...etc. Please note that the region mentioned here are actually a set of voxels stored by their coordinates.

Modifying regions

You may modify a region using [Regions][Modify Regions]. The modification includes shitting the region in x, y, or z-direction. Flip x, flip y, or flip z correct the orientation problem. There is also an expansion function that expands the current region. Other functions include erosion, smoothing, negate, and defragment.

To perform ROI1+ROI2, use [Region][Merge all].

To perform ROI1-ROI2, first "negate" ROI1, and then merge negated ROI1 with ROI2, and then negate the merged regions.

Exporting quantitative index in the region

Once you draw an ROI or ROA, you may export the quantitative indices (e.g. FA, ADC, ...etc) inside the regions by selecting [Regions][Save Region Data As]. The exported information is a text file including the coordinates and the corresponding values for the indices

Randomize Seeding

If "randomize seeding" is off (default setting), DSI Studio will use a "constant" random generator to place the seeds. This means that the seeding sequence will be "deterministic", and each tracking round will generate identical results even though the seeding location is randomly chosen from the seeding regions. This setting ensures that the tracking result is exactly reproducible using the same tracking parameters. 

If "randomize seeding" is on, DSI Studio will use a time variable to initiate the random generate, and the tracking result will be different for every repeated round.

This randomization setting also affects the random selection of "anisotropy threshold", "step size", and "angular threshold". In other words, if "randomize seeding" is off, all random generator will be deterministically-random.

Region Types

There are several region types available to control fiber tracking, including seed, roi, roa, end, terminative. Each of them is explained in the following sections.

My recommendation is to use only ROIs (e.g. one ROI or two ROIs) for tracking, and there is no need to assign any seed regions because DSI Studio will use the whole white matter region as the default seed region. Once a rough track pathway is tracked, a new seed region can be created from the tracks to get more efficient tracking results. This can be done by using [Tracts][Tract to ROI] function, followed by [Regions][Modify Region][Dilation]. 

ROA can be used to eliminate false tracks, whereas the "end" region and "terminative" regions are rarely used. 

An "end" region is more restrictive than an "ROI". An "ROI" allows tracks to pass through, whereas the end region requires that the trajectories end in them. If you cannot find connections using two end regions, try setting them as ROI.


The tracking algorithm will place starting points within seeding "voxels" at subvoxel resolution. The actual seeding points are "uniformly distributed" within the seeding volume. For example, a seed voxel placed at (53,87,68) will have subvoxel seeding placed within (52.5~53.5, 86.5~87.5, 67.5~68.5). Within the voxel region (52.5~53.5, 86.5~87.5, 67.5~68.5), DSI Studio draws a point within the voxel range using a uniform distribution. The point is then used as the starting point within the selected voxel.

Users can specify a seeding point file to override the subvoxel seeding routine and guide the tracking algorithm to start at specific points. To do this, in the tracking parameter, assign "Voxel Center" to the "Seed Position" item and assign "Primary" to the "Seed Orientation". A text file storing a list of point coordinates is needed. For example, to start racking at (53.42, 87.34, 68.43), (53.41, 87.32, 68.32), and (53.67, 87.21, 68.21), you need to have a text file with the following content:

5342 8734 6843
5341 8732 6832
5367 8721 6821
100 -1 -1

Please note that the coordinates are scaled by 100. The largest number accepted is 32767. The number of points will determine the number of tracks generated (In tracking parameters, please make sure that "Terminat if" has a number larger than your point count). 

Open this seed point text file in [Regions][Open region] and change its region type from "ROI" to "Seed". Click "Fiber Tracking" button to start fiber tracking at these points.

If no seed region is assigned, DSI Studio will use the whole brain region as the seed region.

Region of interest (ROI) & region of avoidance (ROA)

ROI is used to "filter in" the tracks that pass through the region, whereas ROAs "filter out" the tracks. Multiple ROI and ROA could be used in combination to fulfill a complicated tracking demand. For example, if there are two ROIs, the tracks that pass through both of the regions are selected. 

End region

An "End" region selects tracts that are ended (not passing) in the "end" region. If one ending region is assigned, then only the tracks ended within the regions are preserved. If two end regions are assigned, then the tracks ended in both regions are preserved. Please note that the "end" region, unlike the terminative region, does not affect the termination of the tracking algorithm. It simply selects the tracts that end in it.

Terminative region

A terminative region terminates any tracts as soon as they enter it. It changes the behavior of a tracking algorithm and force tracking to terminate. A terminative region is useful if one is to study the tracts that project to a nucleus or a specific cortical area. A terminative region does not allow a tract to pass through it, which is very different from an "end" region.

A terminative region can be used to terminate a track if the anisotropy level is greater than a threshold. The steps are the following: (1) In the options window, set the anisotropy threshold to the maximum value (2) click on [Region][Whole Brain seeding]. This creates a region with FA greater than the threshold (3) Change the region type to "terminative" (4) Setting the Fanisotropy threshold back to the minimum value (5) Do fiber tracking.

Step T3c: Options

In the right upper window, there is an options list storing [Tracking Parameters] and other visualization parameters. 

Anisotropy threshold

This parameter determines the threshold for fiber termination. In DTI, FA is used as the index for the filter to determine the fiber threshold. In DSI, QBI, GQI, the fiber threshold is based on the quantitative anisotropy (QA), which is defined for each resolved fiber orientation. The definition of QA is documented in the GQI paper [3].

It is noteworthy that the anisotropy threshold is used as a mask to filter out background voxels. As a result, the estimated along-track FA/QA index estimated will be affected by the fiber threshold.

The initial value of this threshold is determined automatically using 0.6 * (Otsu's threshold). Otsu's method calculates the optimal separation threshold that maximizes the variance between the background and foreground. 

If a threshold of 0 is assigned, DSI Studio will use a randomly selected threshold from [0.5 Otsu, 0.7 Otsu].

The general principle for choosing the threshold is to select the lowest value that has acceptable spurious fibers.

Max angle

This threshold serves as a termination criterion. If two consecutive moving directions have crossing angle above this threshold, the tracking will be terminated.

Assign 0 to do a random selection from 15 degrees to 90 degrees.

Step size

Step size defines the moving distance in each tracking interval. This unit is in millimeter scale. The default setting is the half of the spatial resolution.

Assign 0 to do a random selection of the step size from 0.5 voxel to 1.5 voxel distance.


Smoothing serves like a "momentum". For example, if smoothing is 0, the propagation direction is independent of the previous incoming direction. If the smoothing is 0.5, each moving direction remains 50% of the "momentum", which is the previous propagation vector. This function makes the tracks appeared smoother. In implementation detail, there is a weighting sum on every two consecutive moving directions. For smoothing value 0.2, each subsequent direction has 0.2 weighting contributed from the previous moving direction and 0.8 contributed from the income direction. To disable smoothing set its value to 0.

Assign 1.0 to do a random selection of the value from 0% to 95%.

Min and Max Length Constraint

Length Constraint filters out the tracks that are either too short or too long.

Topology-Informed Pruning (TIP) iterations

Setting it to 1 or 2 to remove false connections using TIP method (Yeh Neurotherapeutics 2018)[5].

Primary orientation versus all orientations 

Specify the starting orientation of the seeds. "Primary orientation" starts the tracking from the primary fiber of a seeding point, whereas "all orientation" may start from a branching fiber.

The "primary" approach always starts the tracking from then most prominent fiber in a voxel. The advantage of "primary" is the stableness and consistency of the results. However, it is possible that the tracks that you are interested in are minor branches and may not be successfully tracked using "primary" setting. A way to compensate for this drawback is using whole brain seeding to explore all possible connections.

The "random" approach starts the tracking a randomly selected fiber orientation, so the results have random factor. The advantage is that "random" can explore all possibility, but the drawback is that the results may not be reproduced exactly. The tracking results are also sensitive to noisy fibers because the a false fiber orientation can be selected as the starting direction.

The "all" approach starts a track for each fiber orientation resolved in a voxel. It aims to explore all possible connection and there are no stochastic factors that may hurt the reproducibility. The drawback is that "all" setting is sensitive to noisy fibers because all resolved fiber orientations will be used as the starting directions.

Subvoxel versus voxel

Specify the seeding strategy. "Subvoxel" conducts subvoxel seeding and each seed voxel may have infinite seeding locations within the voxel. "Voxel" places the seeding location at the center of a seed voxel.


The interpolation method used in estimating the fiber orientation.

Tracking Method

The deterministic fiber tracking method [4] is the default method for fiber tracking. Runge-Kutta method[1] is a higher order tracking method similar to the default Euler approach. 

5000 Tracts v.s. 5000 Seeds

This tracking plan determines when the tracking should stop. The default options are seeding until the total amount of fibers reaches the assigned number. Another option is to have a fixed number of seeds placed in the seeding area. One should note that the generated fibers may be less than the seeds because 
some fibers will be discarded due to length constraint, ROI, ROA, and End point settings.

Thread Count

DSI Studio supports multithread fiber tracking, which can boost the performance on a computer with multiple core CPU. Assign the thread count in accord with the possible computation power to obtain the highest efficiency. 

Check ending

It checks whether the tracking terminates with a location that has anisotropy higher than the threshold. If yes, the fiber is probably terminated in the white matter and will be removed.

Step T3d: Fiber tracking

Click the start tracking button to perform tracking, and input the tract count or the seeding number. The fiber trajectories can be saved using [Tracts][Save Tracts As]. One should note that [Tracts][Save Tracts As] saves only the current selected fiber bundles. To save tracts of different clusters at one, use [Tracts][Save All Tracts As...].

DSI Studio saves tracks in the native diffusion voxel space rotated to "LPS". The coordinates are voxel coordinates started from (0,0,0) at the most left/anterior/bottom point of the image volume. The orientation is (+right,+posterior, +top). For example, (1,2,3) = [the left most 1st voxel, the most anterior second voxel, the bottom 3rd voxel].

To track trajectories connecting between two brain regions, place two "end" regions, one on each of them. DSI Studio will find the trajectories that end in these two regions.

To force track to terminate if the anisotropy threshold is "greater" than a value. First, set up the threshold to the designated value and place whole brain seed by [Regions][Whole Brain Seeding]. In the region list window, change the region type from "Seed" to "Terminative". This will enforce a termination if tracks enter the region. Adjust the threshold to a lower value to initiate fiber tracking.

Step T3e: Track-specific analysis

After fiber tracking, you may use [Tracts][Statistics] to get quantitative metrics related to the tracks. The metrics include

number of tracts       This is the number of streamlines generated by the algorithm
tract length               This is calculated by multiplying number of coordinates in the streamlines with the distance between the coordinates (step size)
tracts volume            This is calculated by multiplying number of voxels passed by all streamlines with the voxel size (in mm cubic)
qa mean                    This is the mean of qa values sampled at all coordinates of the track streamlines.

You can insert an external volume (e.g., T1W, T2W, PET...etc.) using [Slices][Insert T1W/T2W...] and use track statistics to sample them along the fiber pathways.                            

Automatic fiber tracking using atlas-guided track recognition

DSI Studio provides an automatic fiber tracking function that allows users to do a simple one-click for tracking different pathways. The function uses a track recognitioiin based on the a tractography atlas (Yeh et al. 2018) to filter out false tracks and unrelated tracks. 

On the right-hand side, click on the [Enable auto track...] to bring up this function. DSI Studio will normalize the subject's QA map to the MNI QA map and allows track recognition based on the bring tractography atlas. Select the target tracks and click fiber tracking to get the results. 

The detail of how it works is the following:

(1) non-linear registration of subject data to MNI space, (2) seeds are placed within the HCP tractography atlas tract volume (i.e. in any voxel corresponding to any tract), (3) each of the generated streamlines is compared to the streamlines associated with each fiber tract from the HCP tractography atlas (Yeh 2018) using Hausdorff distances, (4) determine which of the generated streamlines are the best match to the streamlines from the target structures in the HCP tractography atlas, (5) streamlines matching to the target track in the HCP tractography atlas are retained and all other streamlines are discarded.

Tractography file formats

The supported file formats include TracVis's TRK format (default), a text file format, or a MATLAB MAT V4 format: 

TrackVis TRK format

The specification for the TRK format is detailed on the TrackVis website. DSI Studio will always output TRK file in LPS convention.

Text trajectory format

The output format can be a text file that stores the coordinates of each fiber track. The coordinates of each fiber trajectory are stored in one line. The x y z coordinates are list sequentially:

x1 y1 z1 x2 y2 z2 ... xn yn zn               <---first tract
x1 y1 z1 x2 y2 z2 ... xn yn zn               <---second


The trajectories can be saved as a Matlab MAT version 4 format. The coordinates of all tracts are stored in a matrix named "tracts". The numbers of coordinates for each fiber are stored in a matrix named "length". The coordinates of the first trajectory are stored in tracts(:,1:length(1)), and the second in tracts(:,length(1)+1,length(1)+length(2)).

To save the tracts data in MAT version 4 format. Use the command save tract.mat -v4

Save Tracks in Native Space

The QSDR reconstructed FIB files allow for tracking in the template space. To save the tracks back to the native space, in step2 reconstruction, you may need to check "Spatial mapping" in the advance option before QSDR reconstruction. This will ensure an output of spatial mapping to the FIB file that allows converting tracks back to the native space.

Once a track is generated in the QSDR FIB file, use [Tracts][Save Tracts][Save Tracts in Native Space] to convert track coordinates to the native diffusion space.

Shortcuts and Graphic Control

Move slices or regions

Right double click in the ROI window to move slices to the pointed location.
Press middle mouse button to drag a slice or a region in the ROI window to the left.
Press Ctrl+M to drag a slice or a region in the 3D window.

Key "Q" and "A": move sagittal slide
Key "W" and "S": move coronal slide
Key "E" and "D": move axial slide

Slice views

Click on the brain buttons at the bottom row of the ROI window to switch between difference slice views
Key "Z": switch to sagittal view
Key "X": switch to coronal view
Key "C": switch to axial view
Use mouse wheel to zoom in or zoom out in the left ROI window

Select a region

Left double click on a region: select it in the region list

Browse 3D objects

Mouse left button: press and rotate the object
Mouse right button: press and change the zoom scale
Mouse mid button: press and move the object (you may also use direction keys to move the object)
Use mouse wheel to zoom in or zoom out in the 3D window

3D views

Alt+1: remember the current viewport and slice position to memory slot 1
1: return to the viewport and slice position recorded in memory slot 1
The same function applies to Alt+2,...Alt+9, and 2,3,...9

Track editing

The following four shortcuts are for track editing. To edit the tracts, 1) hit the shortcut 2) press left mouse button 3) drag the cursor 4) release the mouse button. If the track selection further considers the income angle, use right mouse button instead. (please refer to here for details)

Ctrl+S: select tracts in the 3D window 
Ctrl+D: delete tracts in the 3D window
Ctrl+P: delete tracts in the 3D window
Ctrl+X: cut tracts in the 3D window (click-drag-release)(cannot undo)

Ctrl+T: trim tracts
Ctrl+Z: undo select and delete
Ctrl+Y: redo select and delete

Erase regions in the 3D window

1. Create a cubic region (which will be used as an "eraser" region) and in the region list, move it to the top
2. In the 3D window, use Ctrl+A to drag the eraser region to regions to be erased
3. Use [Regions][Modified Checked Regions][All Excludes First] or Ctrl+Shift+2 to apply the erasing effect
4. Repeat 2 and 3 until the editing is done.

Program Registry

The parameters are stored in the following directory:

On Unix systems, the following files are used by default:

$HOME/.config/DSI Studio.conf
/etc/xdg/DSI Studio.conf

On Mac OS X versions 10.2 and 10.3, these files are used by default:
$HOME/Library/Preferences/com.labsolver.DSI Studio.plist
On Windows, settings are stored in the following registry paths:
HKEY_CURRENT_USER\Software\Labsolver\DSI Studio

Frequently Ask Questions

Q: Why did I get different tracking results each time even though all settings and parameters remained the same?
A: The default settings for angular threshold, anisotropy threshold, and step size use randomized sampling to get values from a working range (see above for details). Manually changing them to a fixed value does not really address the reproducibility problem unless you really need identical results.

A better strategy is "parameter saturation" (Yeh 2020 Neuroimage) which combines randomized parameters with millions of tract or seed count to saturate the parameter space until the results converge. This will improve the test-retest reliability of the analysis and make results robust to different parameter settings.


[1] Basser, P.J., Pajevic, S., Pierpaoli, C., Duda, J., Aldroubi, A., 2000. In vivo fiber tractography using DT-MRI data. Magn Reson Med 44, 625-632.
[2] Wedeen, V.J., Wang, R.P., Schmahmann, J.D., Benner, T., Tseng, W.Y., Dai, G., Pandya, D.N., Hagmann, P., D'Arceuil, H., de Crespigny, A.J., 2008. Diffusion spectrum magnetic resonance imaging (DSI) tractography of crossing fibers. Neuroimage 41, 1267-1277.
[3] Yeh FC, Wedeen VJ, Tseng WY. Generalized q-sampling imaging. IEEE Trans Med Imaging 2010;29:1626-1635.
[4] F.-C. Yeh, T. D. Verstynen, Y. Wang, J. C. Fernández-Miranda, and W-Y. I. Tseng, "Deterministic Diffusion Fiber Tracking Improved By Quantitative Anisotropy", PLoS ONE 8(11): e80713. doi:10.1371/journal.pone.0080713 (pdf)
[5] Yeh, F. C., Panesar, S., Barrios, J., Fernandes, D., Abhinav, K., Meola, A., & Fernandez-Miranda, J. C. (2018). Improved Accuracy of Diffusion MRI Tractography Using Topology-Informed Pruning (TIP). Neurotherapeutics .