Inquiry about BraTS2021 Implementation Strategy in Medical SAM 2

[ObliviateHe]

Dear author

Could you please provide guidance on the correct implementation approach for BraTS2021, or share the specific code/configuration used in your experiments? This would greatly help the research community reproduce your excellent results. Specific questions are as follows:

• Did you use multi-target training (training ET, TC, and WT regions separately) or single-target training (training only whole tumor)?
• Did you generate one bounding box per slice for the whole tumor, or multiple bounding boxes (one for each region: ET, TC, WT)?
• During training, did you use the original BraTS2021 labels (0,1,2,4) directly, or did you convert them to binary masks for each region?

Best
Will

[ncp-nice]

Hello, I would like to ask a question regarding the released source code.
I noticed that the repository only contains code for training and validation, but I could not find a dedicated script for testing on the test set.

May I ask whether the evaluation code used to report the test results on BTCV (i.e., printing Dice scores for all 12 organs individually) and REFUGE was implemented separately by yourself? Or is it included in the released codebase but located elsewhere?