If you have ever worked with microscopy images from different collections, you almost certainly saw that there is a stark difference between them. What is causing this shift? A number of factors are possible: The staining might be different or also the handling of tissue. But one thing that you might not think about at first is the scanner domain shift, introduced solely by different optical properties of the acquisition device.
In this example you see the same tissue (breast tumor), scanned with two scanners of the same manufacturer:
When you train a machine learning model solely on images of the scanner in the left image, then you will very likely not get a solution that generalizes also to the scanner on the right.
To develop approaches for this is the goal of this challenge. We hope we caught your interest and you want to participate now. 🙂
Leave a Reply