This App implements both the TRAINING and TEST PHASES of Classifyber, a supervised streamline-based method that performs automatic bundle segmentation by learning from example bundles already segmented. Classifyber is based on binary linear classification, which simultaneously combines information from bundle geometries, connectivity patterns, and atlases. In addition, it is robust to a multitude of diverse settings, i.e. it can deal with different bundle sizes, tracking algorithms, and dMRI data qualities.
If you want to run only the TEST PHASE, please refer to the App https://doi.org/10.25663/brainlife.app.265.
- Giulia Bertò ([email protected])
- Emanuele Olivetti ([email protected])
We kindly ask that you acknowledge the funding below in your publications and code reusing this code.
We kindly ask that you cite the following article when publishing papers and code using this code:
"Classifyber, a robust streamline-based linear classifier for white matter bundle segmentation", Bertò, G., Bullock, D., Astolfi, P., Hayashi, S., Zigiotto, L., Annicchiarico, L., Corsini, F., De Benedictis, A., Sarubbo, S., Pestilli, F., Avesani, P., Olivetti, E. NeuroImage (2020).
On Brainlife.io
You can submit this App online at https://doi.org/10.25663/brainlife.app.228 via the “Execute” tab.
Inputs:
To perform the bundle segmentation, you need two key elements: (i) the tractogram of the (target) subject you want to extract the bundle from and (ii) the wmc segmentations of multiple (example) subjects you want to learn from. Moreover, you have to provide the anatomical T1s and the tractograms of the (example) subjects (which are used to compute bundle superset and internal conversions). WARNING: all the tractograms need to be already co-registered in the same anatomical space (you can use the App https://doi.org/10.25663/brainlife.app.202 to warp your .tck file to the MNI152 T1 space).
Output:
You will get the wmc segmentation of the bundle(s) of interest in the target subject. You can convert it in multiple .tck files with the App https://doi.org/10.25663/brainlife.app.251.
The wmc segmentation files you have to provide as examples should be obtained using the AFQ segmentation algorithm (https://doi.org/10.25663/brainlife.app.207) or the WMA segmentation algorithm (https://doi.org/10.25663/brainlife.app.188).
In the first case, you can choose the bundle(s) to be segmented by providing the id(s) related to the AFQ segmentation as follows:
1 - Left Thalamic Radiation
2 - Right Thalamic Radiation
3 - Left Corticospinal
4 - Right Corticospinal
5 - Left Cingulum Cingulate
6 - Right Cingulum Cingulate
7 - Left Cingulum Hippocampus
8 - Right Cingulum Hippocampus
9 - Callosum Forceps Major
10 - Callosum Forceps Minor
11 - Left IFOF
12 - Right IFOF
13 - Left ILF
14 - Right ILF
15 - Left SLF
16 - Right SLF
17 - Left Uncinate
18 - Right Uncinate
19 - Left Arcuate
20 - Right Arcuate
In the second case, you can choose the bundle(s) to be segmented by providing the id(s) related to the WMA segmentation as follows:
38 - Left pArc
39 - Right pArc
40 - Left TP-SPL
41 - Right TP-SPL
42 - Left MdLF-SPL
43 - Right MdLF-SPL
44 - Left MdLF-Ang
45 - Right MdLF-Ang
- git clone this repo.
- Inside the cloned directory, create
config.json
with something like the following content with paths to your input files:
{
"tractogram_static": "./track.tck",
"t1_static": "./t1.nii.gz",
"segmentations": [
"./sub-1/classification.mat",
"./sub-2/classification.mat"
],
"tracts": [
"./sub-1/tracts",
"./sub-2/tracts"
],
"tractograms_moving": [
"./sub-1/track.tck",
"./sub-2/track.tck"
],
"t1s_moving": [
"./sub-1/t1.nii.gz",
"./sub-2/t1.nii.gz"
],
"tractID_list": "11, 12, 19, 20"
}
- Launch the App by executing
main
.
./main
This App only requires singularity to run.