I am looking for a large image dataset >20K images to be used for cancer classification algorithm.

Where can I look given that all publicly available datasets are maximum 1K in size which is much less than what I need?

Image resolution does not matter as well as body part as long as all the images are of same resolution and body part.

Images should have benign/malignant labels as ground truth.

Dataset should be half malignant and other half benign or close match.


1 Answer 1


The Cancer Imaging Archive contains a large amount of resources for this exact purpose. While only one dataset has over 20.000 images, you add to this dataset using the multitude of smaller ones listed on the page.

Note that access to the data may be limited in some instances due to the medical nature.

  • 1
    I noticed images are in dicom format, I don't know how to convert into jpg format for DL algorithms. Also data is not labeled, I was looking for something that is labeled like benign/malignant tumor for example.
    – s.ali
    Feb 16, 2019 at 15:05
  • 1
    Stackoverflow has the answer to your first problem: stackoverflow.com/questions/46968678/… It's best to select datasets that have clinical supporting data mentioned in the table. These are the labels that can be downloaded separately.
    – Stephan Heijl
    Feb 16, 2019 at 15:11
  • 1
    Thank you but I did not find what I need on cancer imaging archive as the clinical supporting data contains information like : medicine/patient age ..etc and not something that I can use for classification. I was looking for something more like tumor type ground truth.
    – s.ali
    Feb 18, 2019 at 8:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.