Today we got the confirmation that the MIDOG challenge will be supported by Medical Data Donors (MDD). MDD is a nonprofit organization founded to facilitate anonymous donation of medical data. We share the mindset that medical data needs to be accessible to researchers without barriers to promote creation of next generation AI solutions. The MIDOG organizers, being a team of researchers from different countries, are especially grateful for the support that will enable us to provide awards to the top ranked scientists of the challenge.
Hello, I have some question about non-mitosis cell in test dataset.
1. Is there any non-mitosis cell in test dataset?
2. Is a detected object considered to be a false positive if the Euclidean distance to a non-mitosis cell is less than 7.5 microns?
Looking forward to your reply, thanks!
Hey,
the non-mitosis cells are only hard negatives, i.e. it is not the task to detect these. Your solution should detect only mitotic figures, and the non-mitotic annotations can be utilized for your training scheme (e.g., to achieve stronger gradients by sampling hard negatives). The notebook might be a bit confusion to that end, because non-mitotic cells are also being detected there. Thus: Non-mitotic cells do not play any role in the test set.
W.r.t. your second question: Non mitotic cells are not part of the evaluation. So it is false positive if the Euclidean distance is > 7.5 microns to any real mitosis.
Hope that clarifies everything 🙂
Best,
Marc