Skip to content

Latest commit

 

History

History
114 lines (88 loc) · 6.16 KB

README.md

File metadata and controls

114 lines (88 loc) · 6.16 KB

Enhanced short fiber bundle segmentation

fs

This repository contains the code used in: "Short fiber bundle filtering and test-retest reproducibility of the superficial white matter". Which enables the segmentation of well-defined Superficial White Matter (SWM) fiber bundles.

Four fiber bundle filters are available to remove spurious fibers from an automatic segmentation method [1] based on a SWM multi-subject atlas [2]. The filtering post-processing step helps to produce smoother bundles with less isolated fibers. Aditionally, the main fiber fascicle (MFF) identification helps to identify well-defined short fiber bundles. Available fiber bundle filters are based on:

  • Connectivity Patterns.
  • Symmetric Segment-Path Distance.
  • Fiber Consistency.
  • Convex Hull (Recommended)

The enhanced short fiber bundle segmentation can be done in two ways:

  • Fiber bundle segmentation + Fiber bundle filter is applied.
  • Fiber bundle segmentation + Main fiber fascicle identification + Fiber bundle filter is applied.

Also, a folder with segmented fiber bundles can be provided to apply a fiber bundle filter with custom parameters.

Dependencies

  • C++ compiler
  • Ubuntu 18.04.6 LTS
  • OpenMP >= 4.5
  • Python >= 3.6
  • Dipy >= 1.4.1
  • Numpy >= 1.21.5
  • Scipy >= 1.7.3
  • Scikit-learn >= 1.0.2
  • Nibabel >= 3.2.2
  • Joblib >= 1.1.0

Quick install

git clone https://github.com/cmendosanchez/EnhancedSWM.git

Use example

Fiber bundle segmentation + Fiber bundle filter is applied.

python3 main.py --in_data input --extension extension --out_dir out_dir --filter filter_number
  • --in_data input tractogram in MNI space (tck/trk/bundles)
  • --extension extension of tractogram (tck/trk/bundles)
  • --out_dir name of the folder containing the segmented fiber bundles
  • --filter integer defining a fiber bundle filter (1: Connectivity Pattern, 2: SSPD, 3: Fiber Consistency, 4: Convex Hull)

Fiber bundle segmentation + Fiber bundle filter is applied & Main fiber fascicle identification + Fiber bundle filter is applied.

python3 main.py --in_data input --extension extension --out_dir out_dir --filter filter_number --MFF 1
  • --in_data input tractogram in MNI space (tck/trk/bundles)
  • --extension extension of tractogram (tck/trk/bundles)
  • --out_dir name of the folder containing the output files
  • --filter integer defining a fiber bundle filter (1: Connectivity Pattern, 2: SSPD, 3: Fiber Consistency, 4: Convex Hull)
  • --MFF apply main fiber fascicle identification and fiber bundle filtering. 1: True 0: False (default False)

Execution with example tractogram

An example tractogram in MNI space can be download from here (available in tck/trk/bundles format).

python3 main.py --in_data Data/Tractogram_MNI.tck --extension tck --out_dir Ex_seg/Seg --filter 4 --MFF 1

Resulting folders

  • --out_dir folder containing the segmented short fiber bundles
  • Filtered_bundles_${selected filter}_${format tck/trk/} folder contatining the filtered short fiber bundles. By default a folder with .bundles format is generated for I/O operations.
  • MFF folder containing the main fiber fascicles (if --MFF 1)
  • Filtered_MFF_${selected filter} folder contatining the filtered main fiber fascicles (if --MFF 1)

If the indices of the fibers with respect to the orignal tractogram are needed, these are provided in the respective '_idx' folder. Resulting fiber bundles have 21 equidistand points.

Fiber bundle filter is applied to a folder of fiber bundles with custom parameters.

Fiber bundles must have 21 points, if your data is in .tck you can use tckresample (https://mrtrix.readthedocs.io/en/3.0.4/reference/commands/tckresample.html) or dipy set_number_of_points function (https://dipy.org/documentation/1.4.1./reference/dipy.tracking/#set-number-of-points).

python3 main.py --in_folder input_folder --extension --filter filter_number --p1 --p2
  • --in_folder input folder with fiber bundles in MNI space (tck/trk/bundles) (bundles must be in MNI to perform data format conversion and I/O operations)
  • --extension extension of the fiber bundles (tck/trk/bundles)
  • --filter integer defining a fiber bundle filter (1: Connectivity Pattern, 2: SSPD, 3: Fiber Consistency, 4: Convex Hull)
  • --p1 first parameter of the fiber bundle filter (percentage of discarded fibers)
  • --p2 second parameter of the fiber bundle filter (1: $\theta_{END}$, 2: $\theta_{SSPD}$ 3: $K_{f}$, 4: $K_{p}$)

Execution with example folder of short fiber bundles

An example folder with short fiber bundle can be download from here (available in tck/trk/bundles format):

python3 main.py --in_folder Bundles/Seg --extension tck --filter 4 --p1 20 --p2 80

will apply the fiber bundle filter based on the Convex Hull, discarding 20% of the fibers and using $K_{p}=80$.

Performance

The time (minutes) to perform the short fiber bundle segmentation in a tractogram of 3 millions steamlines, using a desktop computer with an AMD Ryzen 9 5900X 12-Core processor. If data is provided in .tck or .trk file format, then it could take longer due to data conversion to .bundles format and I/O operations. Due to the high computational cost of the SSPD distance, the filtering of a high number of fiber bundles could take several hours.

MFF = 0 MFF = 1
Connectivity Patterns 2.50 2.84
SSPD - -
Fiber Consistency 3.86 5.64
Convex Hull 3.14 3.68

References

[1] A. Vázquez, N. López-López, N. Labra, M. Figueroa, C. Poupon, J.-F. Mangin, C. Hernández, and P. Guevara, “Parallel optimization of fiber bundle segmentation for massive tractography datasets,” in 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019). IEEE, apr 2019.

[2] C. Román, C. Hernández, M. Figueroa, J. Houenou, C. Poupon, J.-F. Mangin, and P. Guevara, “Superficial white matter bundle atlas based on hierarchical fiber clustering over probabilistic tractography data,” NeuroImage, vol. 262, p. 119550, nov 2022.