Recommended datasets for transfer learning

[Ondřej Plátek]

Hi @ieee8023

thank you for maintaining this dataset!

I implemented a pytorch lightning wrapper for a DenseNet model for covid-chestxray-dataset.

It is kick of a Pytorch Lightning's community project which aims at to be covid19 detector (for educational purposes).

Can you recommend us datasets and strategies on how to use additional data

I have scanned https://arxiv.org/pdf/2002.02497.pdf (I will return to it). It seems that to solve the labeling differences and other dataset preparation differences quite a lot of domain expertise is needed. Any tips appreciated.

Kind regards

Ondra

PS: I was inspired by #15
PPS: My fork was merged to the PyTorchLightning community project
PPPS: I believe that @Borda already contacted you that we may use slack for longer discussions if needed. Link to the slack can be found at PL

[Paul Kienzle]

Not a recommendation, but here are some links to datasets that I found:

• NIH chest X-ray data
  • https://nihcc.app.box.com/v/ChestXray-NIHCC/file/219760887468
  • 112k images
• Kaggle copy of NIH dataset
  • https://www.kaggle.com/nih-chest-xrays/data
• Kaggle dataset for pneumonia in children
  • https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia
  • 2538 bacterial, 1345 viral and 1349 normal
• Stanford dataset
  • https://stanfordmlgroup.github.io/competitions/chexpert/
  • 224316 images from Stanford hospital, 16627 no finding
• PadChest
  • http://bimcv.cipf.es/bimcv-projects/padchest/
  • 160k images from Valencia databank, many with manual labelling
• NIH Tuberculosis collection
  • https://openi.nlm.nih.gov/faq#faq-tb-coll (Open the "[+] I have heard about the Tuberculosis collection.")
  • Montgomery County X-ray set: 80 normal, 58 abnormal
  • Shenzhen set: 340 normal, 275 abnormal
• NIH open-i Indiana University collection:
  • https://openi.nlm.nih.gov/faq#faq-tb-coll (Open the "[+] Where can I get the Chest X-ray images in Open-i?")
  • 8121 images
• MIMIC-CXR
  • https://physionet.org/content/mimic-cxr/2.0.0/
  • 377k DICOM images from BIDMC in Boston
  • Need to be a member of PhysioNet to access

I also found a list of public medical imaging data collections, not restricted to lung X-ray at https://www.radrounds.com/profiles/blogs/list-of-open-access-medical-imaging-datasets
radrounds.com

[Ammar Chalifah]

Hi Oplatek and Pkienzle, thank you so much for the question and the answer! Currently I'm trying to train a covid chest x-ray detection model. I have trained the classifier with Kaggle dataset. Now I want to train the classifier using covid, pneumonia, and normal data. I think I want to use covid data from this repository, but I'm having difficulty to find normal data. Are there any reference or suggestion?

Thank you so much!

[Joseph Paul Cohen]

Please check out this paper for a transfer learning approach and tasks to work on (and the clinical workflows that could benefit from tools): http://arxiv.org/abs/2006.11988

Also check out this library for dataloaders for over 7 different datasets: https://github.com/mlmed/torchxrayvision