This README file describes the models proposed in the current folder.

The models were created using clinicadl == 1.0.4. Each folder containing the
models is compressed in a tar.gz file. The filename corresponds to the
experiments described in the supplementary material of the main publication
[1], see the eTable 4.

Here a simplified version of the aforementioned table:

| Experiment |     Architecture      | Training Data | Transfer learning |   Task   |
|     3      |  3D subject-level CNN |    Baseline   |        AE         | AD vs CN |
|     8      |  3D roi-based CNN     |    Baseline   |        AE         | AD vs CN |
|     14     |  3D patch-level CNN   |    Baseline   |        AE         | AD vs CN |
|     18     |  2D slice-level CNN   |    Baseline   | ImageNet pretrain | AD vs CN |

Model architecture, weights and hyperparameters are self-contained in each
folder and are organized by followint the MAPS structure [2].

[1] Junhao Wen, Elina Thibeau-Sutre, Mauricio Diaz-Melo, Jorge Samper-González,
Alexandre Routier, Simona Bottani, Didier Dormont, Stanley Durrleman, Ninon
Burgos, Olivier Colliot, Convolutional neural networks for classification of
Alzheimer's disease: Overview and reproducible evaluation, Medical Image
Analysis, Volume 63, 2020, 101694, ISSN 1361-8415.

[2] https://clinicadl.readthedocs.io/en/stable/Introduction/#maps-definition

@Copyright 2020-2022, Aramislab, Inria.