With the publication of FRAPPE paper (https://doi.org/10.1002/mrm.28237) and Polar segmentation paper (DOI: 10.1109/ACCESS.2020.3040616), we have uploaded the popliteal manual contours for development of FRAPPE. 

https://github.com/clatfd/OAI-Polar

Due to the data sharing policy, we cannot post original images here. Please acquire the images from https://nda.nih.gov/oai/ 

There are four datasets posted here. Train and validation sets are used for developing FRAPPE from the baseline scans in the OAI dataset. After development, Rep set is used to evaluate the reproducibility between FRAPPE vs human reviewer 1, FRAPPE vs human reviewer 2, and human reviewer 1 vs human reviewer 2. Finally, the test set (half labeled by human reviewer 1 and half by human reivewer 2) is used to report segmentation accuracy. 23 cases for training, 2 for validation, 10 for reproducibility, 25 for testing.

It took about 4 hours for labeling each case, but thanks to FRAPPE, 7 minutes are needed to process each case with similar quality, so that we can process the whole OAI dataset within 2 months.