About ACDC DSC Metric

[zqk975]

Hi, I would like to ask the authors, is the dsc metric with a value of 92.83 in the ACDC dataset in your paper a validation set or is it a test set metric for inference, because when I look at the source code, your test set is treated as a validation set?

[Amshaker]

Hi @QikaiZhang ,

Thanks for the question.

The DSC metric for ACDC dataset (92.83%) is based on the testing set not the validation set. The 100 patients are divided into 70 training samples, 10 validation samples, and 20 test samples. We use the same data division of nnFormer paper. Please note that the testing samples are defined here: https://github.com/Amshaker/unetr_plus_plus/blob/bc6bed1eb2d2d735d5f44740958ef4ea862304a6/unetr_pp/training/network_training/unetr_pp_trainer_acdc.py#L401C1-L402C1

Let me know if you have any questions.

Best regards, Abdelrahman.

[zqk975]

Thank you for your careful answer, but I think there are still some confusing points, in this py script you gave, in this code file, the way you divide your data is 80 for training and 20 for validation, and your validation part overlaps with the test part of nnFormer, 100 data samples you used all for training and validation, so what are your test set? Or give the serial number of these test samples.

Looking forward to your answer.