Dear BraTS team, Thank you for your great effort into preparing training and validation data for all participants. After checking the validation dataset, we have two questions: (1) Does each subject have only one connected region of whole tumor? The subject 'BraTS2021_01792' has clearly two disconnected tumor regions (2 WT) for us, we're not sure if the ground truth includes these two parts. (2) Some of the validation data are of low-quality, with artifacts and tumor is poorly highlighted (especially ET on T1ce) for example, #01719, #01732 etc. (we can provide a full list of cases if needed). Are these "difficult" cases deliberately curated in validation dataset to make the validation data as diverse as possible? Or just some images are of low quality and low contrast on tumor? I am looking forward to hearing from you. Thanks, Zjiang

Created by Zhifan Jiang zjiang
Thank you for checking it!
Hi @volcanofly, Since this case is from the validation cohort, we would refrain from commenting on the details of the ground truth. The annotation for this particular case is appropriate. Please refer to the earlier message in this thread.
Hi @ujjwalbaid I looked at the case BraTS2021_01792, and I was suspecting the ground truth might have missed one tumor. I might be wrong. But I was wondering if it is possible that you could help check it manually. Thank you a lot! Best, Can
Hi @zjiang, Thank you for bringing up this interesting question. 1. If any subject has multiple disconnected tumors then the ground truth will also have multiple tumors annotated. Such cases are common in multifocal tumors. 2. In the BraTS challenge the dataset is collected from multiple data contributors from all over the globe with varying scanners and acquisition protocols. There are few such cases present in the training/validation/testing cohort. This variation would make the approach more generalized.

Query about subjects in the validation dataset page is loading…