hiplay7-thickness

Measure the thickness of the hippocampus from 7T MRI using a diffeomorphic vector field approach

Authors: Émilie Gerardin (ARAMIS Lab), Ana B. Graciano Fouquier (ARAMIS Lab), Alexis Guyot (ARAMIS Lab).

This software tool allows to extract central surface representations along with thickness maps from MR images of the hippocampus at 7 Tesla. More tools and resources to study the hippocampus using 7T MRI are available at http://www.aramislab.fr/sevenhipp/ .

Overview

For any use of this code, please cite the following article:

A Guyot*, AB Graciano Fouquier*, É Gerardin*, M Chupin, JA Glaunès, L Marrakchi-Kacem, J Germain, C Boutet, C Cury, L Hertz-Pannier, A Vignaud, S Durrleman, TR Henry, PF Van de Moortele, A Trouvé, O Colliot. (* denotes equal first authors) 'A Diffeomorphic Vector Field Approach to Analyze the Thickness of the Hippocampus from 7T MRI', IEEE Transactions on Biomedical Engineering, 2020, DOI: 10.1109/TBME.2020.2999941

This article is available at https://hal.inria.fr/hal-02359660 .

Installation

The following software and libraries: - Matlab - Python (version 3.2 or newer) - numpy - scipy - matplotlib - joblib - nibabel - nilearn - vtk - Deformetrica (version 4.2.0)

In case you are not sure you already have the relevant Python libraries (numpy, scipy, matplotlib, joblib, nibabel, nilearn, vtk), we recommend installing Miniconda, a program that lets you install and run Python packages and their dependencies into local, user-defined environments.

Miniconda can be obtained at the following website: https://docs.conda.io/en/latest/miniconda.html. Please make sure you choose the version corresponding to your operating system (Windows, Mac OS X or Linux) and to the architecture of your computer (32bit or 64bit).

Deformetrica (http://www.deformetrica.org/) can be installed on a Conda environment. See https://gitlab.com/icm-institute/aramislab/deformetrica for more information.

Usage

Extract the archive 'inputData.tar.gz' present at the root of the git repository with the following command: tar -xvzf inputData.tar.gz This will create a folder 'inputData' at the root of the current directory. You can move the 'inputData' folder wherever you wish, as long as you keep track of the location (which will be referred to as [input_folder] in the following instructions).

Optional: if you have installed dependencies via miniconda, activate the conda environment that contains the dependencies.

To get the results presented in the article, run the following commands:

• Experiment III. A. Robustness with respect to kernel size and anisotropy: python ./launch_experiment_A.py [input_folder] [output_folder_A]
• Experiment III. B. Influence of inter-rater variability: python ./launch_experiment_B.py [input_folder] [output_folder_B] (-n [n_cores])
• Experiment III. C. Application to in vivo 7T MRI group studies: python ./launch_experiment_C.py [input_folder] [output_folder_C] (-n [n_cores])

Where: - [inputfolder] is the path to the extracted archive containing the input data for all the experiments described in the article. - [outputfolderA], [outputfolderB] and [outputfolderC] are separate empty folders where the output data for experiments A, B and C respectively, will be stored. - ncores is an optional argument defining the number of cores to be used in parallel tasks. Default is 1.

Output

Experiment III. A. Robustness with respect to kernel size and anisotropy

outputfolderA will be populated with the following folders:

• '4-centralSurfaceAndThicknessEstimation': contains files for the central surface calculated with a kernel size t=10

• '5-influenceOfKernelSize': contains two subfolders

  • '1-computeMaps': contains files for the central surfaces computed for kernel sizes 3, 5, 10 and 15
  • '2-compareMaps': contains comparisons from the surfaces at kernel size 3, 5, 15 to the surface at kernel size 10 using 3 metrics (thickness correlation, mean abs. thickness difference and mean inter-surface distance).

• '6-influenceOfAnisotropy': contains three subfolders

  • '1-createAnisotropicSegmentations': populated with subfolders 'subample1', ..., 'subsample6', each containing subsampled segmentations for subsampling factors in 1, 2, ..., 6
  • '2-computeMaps': contains files for the central surfaces computed for each subsampled segmentation
  • '3-compareMaps': contains comparisons from the central surfaces obtained with subsampling factors 2, 3, ..., 6 to the central surface obtained with subsampling factor 1

Experiment III. B. Influence of inter-rater variability

outputfolderB will be populated with the following folders:

• '4-centralSurfaceAndThicknessEstimation': contains files for the central surface calculated with a kernel size t=10 for rater1/rater2, subject1/subject2/subject3/subject4 and left/right hippocampus

• '5-interRaterAnalysis': contains two subfolders

  • '1-compareIndivididual': contains the comparison (using the 3 metrics described above) between rater 1 and rater 2 for the left/right hippocampus of subject1/subject2/subject3/subject4
  • '2-compareAverage': contains the three metrics averaged across all four subjects for both left and right hippocampus

Experiment III. C. Application to in vivo 7T MRI group studies

outputfolderC will be populated with the following folders:

• '4-groupStudy': single folder containing five subfolders
  • '1-centralSurfaceAndThicknessEstimation': contains files for the central surfaces calculated with a kernel size of all subjects (controls: control1, control2, control3, control4, control6, control8, control9, control10, control11 ; left contralateral patients: patient1, patient2, patient6 ; left ipsilateral patients: patient3, patient5, patient8, patient9, patient12), for both left and right hippocampus
  • '2-centralSurfaceTemplates': contain the template hippocampus surface computed by Deformetrica for both {controls + contralateral patients} and {controls + ipsilateral patients} groups for both left and right hippocampus
  • '3-templateProjections': for the two groups described above and for both hippocampus, contains the average of all projected thickness maps from controls, the average of all projected thickness maps from ipsi- or contra- lateral patients and the difference map between these two averages
  • '4-avgThicknessVolumeComputation': contains files reporting the volume and average thickness of all hippocampi described in this experiment
  • '5-spearmanCorrelationsComputation': contains file 'spearman_correlations.json' which reports Spearman's rank correlation coefficents between volume and average thickness for the left and right hippocampi (computed across all subjects)

License

With the exceptions of files taken from external libraries, this code is released under the terms of the MIT License. See file LICENSE for further precisions.