search

DIAGNijmegen / picai_labels

Annotations for the PI-CAI Challenge: Public Training and Development Dataset
https://pi-cai.grand-challenge.org/
Other
47 stars 23 forks source link

Ground-Truth missing #12

Closed pouyashiri closed 7 months ago

pouyashiri commented 7 months ago

We are doing research using this data, it seems like ground-truth for the following Radical Prostatectomy cases is missing: 10013 10053 10229 10282 10377 10399 10402 10519 10865 10873 10883 11072

Please let me know if you could help with providing these.

joeranbosma commented 7 months ago

Hi @pouyashiri,

You are correct, these patients indeed received a radical prostatectomy, which came back positive (GS >= 7) (I double-checked the first three).

No manual voxel-level annotations were made for these cases, these are also not available in our archives. The reason for this is the setup of PI-CAI: manually annotating at the voxel-level all 10,000+ cases in the PI-CAI challenge (public + sequestered training data + tuning cohort + testing cohort) is infeasible (we did annotate all at a case-level). For the PI-CAI challenge, we observed that all top-5 solutions used semi-supervised learning to improve performance, with most teams using the AI-derived lesion segmentations provided here: https://github.com/DIAGNijmegen/picai_labels/tree/main/csPCa_lesion_delineations/AI/Bosma22a.

There you can also find the AI-derived annotations for the patients you linked. For more information about how these annotations were derived, please see https://pubs.rsna.org/doi/10.1148/ryai.230031.