Data and models described in 

Reliable learning with PDE-based CNNs and DenseNets fordetecting Covid-19, Pneumonia and Tuberculosis from chestX-Rays images 
- article in review

1.PDE inspired CNNs
parabolic and hyperbolic networks implemented in Meganet/examples/2018-jvmm


pde/data:
i) Exp1 - normal vs pneumonia

It includes all the X-Rays for normal and pneumonia, 0.8 for training

Ctrain_normal.mat, Cval_normal.mat - labels
Ytrain_normal.mat, Yval_normal.mat - RGB images, 192x192

ii) Exp2 and Exp3 - covid, normal, pneumonia, tuberculosis

It includes all Covid-19 AP or PA from Cohen dataset, gathered from Radiopaedia or SIRM. Version from december 2020 of the Cohen dataset 
+ 350 randomly selected normal + 350 randomly selected pneumonia

Ctrain_down.mat, Cval_down.mat - labels
Ytrain_down.mat, Yval_down.mat - RGB images, 192x192

The entire dataset (0.8 training) - higly imbalanced
Ctrain.mat, Cval.mat - labels
Ytrain.mat, Yval.mat - RGB images, 192x192


pde/models
Logs and models for Exp1, Exp2 and Exp3

2. DenseNet 2-Phase
Chexpert_2Phase

It includes 
 i) 2 models for phase 1 - training from CheXpert
ii) 3 models for phase 2 - training for multioutput covid, normal, pneumonia, tuberculosis

3. Progression-regression.pdf includes the results of models CheXpert14 and CheXpert-Covid19 on a case publicly available from Radiopaedie:
https://radiopaedia.org/cases/covid-19-pneumonia-progression-and-regression
This case includes several X-Rays of the same patient during a month. 
Observation: None of the X-Ray included in this test are seen on training the models.