Predict age from brain grey matter (regression). Aging is associated with is grey matter (GM) atrophy, each year, an adult lose 0.1% of GM. We will try to learn a predictor of the chronological age (true age) using GM measurements on the brain on a population of healthy control participants.
Such a predictor provides the expected brain age of a subject. Deviation from this expected brain age indicates acceleration or slowdown of the aging process which may be associated with a pathological neurobiological process or protective factor of aging.
There are 357 samples in the training set and 90 samples in the test set.
Voxel-based_morphometry VBM using cat12 software which provides:
-
Regions Of Interest (
rois) of Grey Matter (GM) scaled for the Total Intracranial Volume (TIV):[train|test]_rois.csv284 features. -
VBM GM 3D maps or images (
vbm3d) of voxels in the MNI space:[train|test]_vbm.npzcontains 3D images of shapes (121, 145, 121). This npz contains the 3D mask and the affine transformation to MNI referential. Masking the brain provide flat 331 695 input features (voxels) for each participant.
By default problem.get_[train|test]_data() return the concatenation of 284 ROIs of
Grey Matter (GM) features with 331 695 features (voxels) within a brain mask.
Those two blocks are higly redundant.
To select only on ROIs (rois) features do:
X[:, :284]
To select only on (vbm) (voxel with the brain) features do:
X[:, 284:]
The target can be found in [test|train]_participants.csv files, selecting the
age column for regression problem.
The main Evaluation metrics is the Root-mean-square deviation RMSE. We will also look at the R-squared R2.
This starting kit requires Python and the following dependencies:
numpyscipypandasscikit-learnmatplolibseabornjupyterramp-workflow
Therefore, we advise you to install Anaconda distribution which include almost all dependencies.
Only nilearn and ramp-workflow are not included by default in the Anaconda
distribution. They will be installed from the execution of the notebook.
To run a submission and the notebook you will need the dependencies listed in requirements.txt. You can install the dependencies with the following command-line:
pip install -U -r requirements.txt
If you are using conda, we provide an environment.yml file for similar usage.
conda env create -f environment.yml
Then, you can activate the environment using:
conda activate brain_age
And desactivate using
conda deactivate
- download the data locally:
python download_data.py
- Execute the jupyter notebook, from the root directory using:
jupyter notebook brain_age_starting_kit.ipynb
Tune your model using the starting_kit
- Submission (Run locally)
The submissions need to be located in the submissions folder.
For instance for starting_kit, it should be located in
submissions/submissions/starting_kit.
Copy everything required to build your estimator in a submission file:
submissions/submissions/starting_kit/estimator.py.
This file must contain a function get_estimator().
Run locally:
ramp-test --submission starting_kit
- Submission on RAMP: