While the decision if MIDOG will be accepted to MICCAI 2021 is not yet there, we ramp up the scanning and annotation process. I explicitly want to highlight the great effort that Nikolas Stathonikos and Natalie ter Hoeve put into this.
It might be considered self-evident, but a challenge can only be good if the dataset is of high quality – and quite a number of challenges in the past have shown that if the proper diligence here is missing, the dataset has severe flaws. To ensure the quality that we expect from all our data sets, Natalie and Nikolas went through all of the 300 cases to make sure that they (1) represent a decent tissue and scanning quality and (2) also are valid breast cancer cases where a mitotic count was clinically evaluated.
On the “other side” of things, Christof Bertram (one of our pathologists) started selecting the ROIs for the mitotic count for the vast majority of images already.
Because he is such a hard worker, he also already started screening the images for mitoses, and completed 2/3 of the training set. To be clear: This is only the first step of annotation, since a second, independent, pathologist will also judge each mitotic figure. And we will also run object detection on each scanner dataset to look for additional candidates, then to be judged again by the pathologists. That’s a lot of work still pending, but it’s worth it for a fine quality data set – and a good challenge!
Thank you all for your hard work in putting this together. It provides the opportunity to solve an important problem in medical imaging today.
Thank you so much for this feedback! We hope that we can provide a high quality data set to the community, and a joyful challenge 🙂