Amshaker / unetr_plus_plus

[IEEE TMI-2024] UNETR++: Delving into Efficient and Accurate 3D Medical Image Segmentation
Apache License 2.0
340 stars 32 forks source link

About synapse dataset labels #30

Closed CamillerFerros closed 1 year ago

CamillerFerros commented 1 year ago

Regarding the synapse dataset, there are a total of 13 organs in the labels in the 3D dataset, but there are a total of 8 organs compared in the paper, please ask whether the function of calculating loss during training uses 8 organs or 13 organs, and whether the model output is 13 organs or 8 organs, thank you very much

Amshaker commented 1 year ago

Hi @CamillerFerros ,

Yes, there are 13 organs and the output of the model is 13 classes, but the DSC and HD95 are computed based on only 8 organs. Please note that the split for Synapse (18 + 12) is different from the split for BTCV (24 + 6).