There are data issues with derived series the following studies in the training set:
ID 760011. Visit T0
ID 861745. Visits T1 and T2
DCE: The "cropped original" 4D DCE series ( Series description = "ISPY2: VOLSER: uni-lateral cropped: original DCE", series ID = n1800) do not have the temporal index field set, and thus will likely not read in correctly. (Sorting by file name will not fix this).
DWI: The MRI parameter DICOM fields (mostly group 0018x; TE, TR etc...) are missing or incorrect in the derived objects (TRACE images; ADC maps).
Replacement series have been generated and are undergoing curation at TCIA. We hope to have them ready for download early in the week of July 19.
We apologize for any inconvenience this might cause, and thanks to Pat Bolan for his careful examination of the data set which caught these issues.
David Newitt, UCSF
THIS NOTICE ONLY APPLIES TO YOU IF YOU DOWNLOADED THE BMMR2 Training Data Set BEFORE July 16th, 2021
As mentioned back in July, it was found that 2 subjects (ID numbers 760011 and 861745) in the BMMR2 training set had derived MR objects with corrupted DICOM header fields. This could result in the multiple-volume images not being sorted correctly in DICOM viewers. On investigation we found that the original DCE images in these studies also had minor header issues, in that many of the MRI parameter fields had been stripped in processing.
If you downloaded the Training Set BEFORE July 16th you may want to download the corrected series. You can just download the entire Training Set again, but since this can be quite time consuming a manifest file for just the corrected series is being emailed to all participants.
- If you make use of the “metadata.csv” files provided with the TCIA download make sure to copy the old one to a new name before doing this new download, as it will be overwritten.
- Downloads on or after July 15th contain the corrected series.
- The Test Set data was not affected by this issue.
Apologies for any inconvenience,