covid-chestxray-dataset icon indicating copy to clipboard operation
covid-chestxray-dataset copied to clipboard

Published 20 hours ago • verified publisher icon ieee8023

Classification example

Open Scitator opened this issue 4 years ago • 3 comments

Hi,

I have prepared an example of Xray classification , based on this repo https://colab.research.google.com/drive/1KlKvHDgvi-cfrpJUIOCvczmL4Ctc1wBL

Please check it out. Are you interested in such initiatives?

Scitator avatar Mar 20 '20 14:03 Scitator

Hi, we prepared DenseNet model using Pytorch Lightning

https://github.com/PyTorchLightning/lightning-Covid19

oplatek avatar Mar 22 '20 14:03 oplatek

@oplatek Several suggestions (just my thoughts during example preparation)

  1. Metadata have a bunch of useful information; not only covid-19, but other disease labels also - you need to use them for better machine learning model
  2. Due to class disbalance, accuracy is not significant metric - use AUC or F1 score instead
  3. Due to low amount of data (only around 100 images) data augmentation is important to prevent overfitting
  4. Because of medicine field, reproducibility really matters it this case; but this is Catalyst duty

long story short, all these DL best practices were implemented during example preparation, https://colab.research.google.com/drive/1KlKvHDgvi-cfrpJUIOCvczmL4Ctc1wBL Make the world a better with better science :)

Scitator avatar Mar 23 '20 12:03 Scitator

@Scitator thank you for the points!

Replacing accuracy is so obvious and it should be there even for the baseline model. -> https://github.com/PyTorchLightning/lightning-Covid19/issues/15

Can you elaborate more on 4? How does Catalyst help reproducibility more than any other framework?

oplatek avatar Mar 23 '20 21:03 oplatek