Using Squeeze-and-Excitation blocks to improve an accuracy of automatic knee osteoarthritis severity grading using convolutional neural networks

First, download and unpack oai224.zip/oai224_4.zip to dataset/ directory.

Dataset for 5 OA grades: https://drive.google.com/file/d/1TyuTBKSjnhcJK4a4DW7Hfs-6sbwXypV5/view?usp=sharing Dataset for 4 OA grades: https://drive.google.com/file/d/1bL5GO6UCj84M9_LwPhAcb97ZEC0rpZ9F/view?usp=sharing

Training (see train.ipynb)

Customize parameters in block Changeable parameters and run.

Testing (see test.ipynb)

Customize parameters in block Choose architecture and model checkpoint and run. You can download all trained models for all used architectures and random seeds by the links:

• classification for 5 OA grades: https://disk.yandex.ru/d/F5feQKmxwHPVxw
• classification for 4 OA grades: https://disk.yandex.ru/d/bIp34P1dD9FMuw

Test ensembles (see test_ensemble.ipynb)

Customize parameters in block Prepare ensemble parts and run.