README.md

ADNI dataset imaging preprocessing

Alzheimer’s Disease Neuroimaging Initiative ADNI T1-weighted MRI pre-processing for deep learning pipelines. Huge thanks to Tian Xia for sharing initial code.

Raw data

Preprocessed data

Histograms

------->

Processing steps

1. Reorientation (FSL fslreorient2std)
2. Cropping (FSL robustfov)
3. Brain Extraction (FSL bet)
4. Atlas Affine Registration (FSL flirt)
   1. Resolution (1 or 2 mm) can be chosen at this step
5. Structure segmentation and bias correction (FSL fast)
6. Intensity normalization between [-1,1]
   1. Clip intensity at 99.5% of max values (numpy.percentile)
7. Central slice cropping. Default to 60mm around the centre along the z axis.

Downloading the data

For downloading the dataset, we refer the user to the ADNI website . You will need to apply for the data with a brief description of your project. Upon acceptance (which should take only a few days), you will download the images through the LONI Image and Data Archive. After login in and selecting the ADNI dataset, you will need to download at least two files:

1. Summary spreadsheet
   1. Download > Study Data > Data&Database > Key ADNI tables merged into one table (file)
2. MRI Images
   1. Download > Image Collections > Other Shared Collections
      1. ADNI1_Screening_1.5T
      2. ADNI1_Complete_1Yr_1.5T
      3. ADNI1_Complete_2Yr_1.5T
      4. ADNI1_Complete_3Yr_1.5T

Unzip the imaging folders and save them in a single folder called "raw_data" which will have the following structure:

raw_data/<subject_id>/<preprocessing>/<date>/<acquisition_id>/<file_name>.nii

Dependencies

Create a conda environment with

conda env create -f environment.yml

then, activate the environment with

conda activate adni

Most image processing methods were done using using the FSL Neuroimaging tool. You may install it by running:

python2 fsl_ubuntu/fslinstaller.py

or by following the instructions here.