Table 3 Comparison of original TCIA format versus BreastDCEDL standardized format, detailing differences in data representation and metadata fields.
From: BreastDCEDL: A standardized deep learning-ready breast DCE-MRI dataset of 2,070 patients
Aspect | Original TCIA | BreastDCEDL |
|---|---|---|
Format | 8.5M DICOM slices (4.6TB) | 11,717 NIfTI volumes (183GB) |
Organization | DICOM files of slices and 3D data, | 3D NIfTI file per acquisition |
| Â | mixed modalities and acquisitions | with standardized naming: |
| Â | with vendor-specific tags | <id>_<ds>_acq<N>.nii.gz |
Temporal | Vendor and clinic-specific | Standardized temporal indices |
alignment | custom tags per slice | Â |
Segmentations | Bitmaps encoding | Binary 3D masks (1,154) + |
| Â | segmentation procedures | Bounding boxes (916) |
Metadata | Embedded in 8.5M DICOM headers | Single unified CSV file |
Load time | Approximately 5–10 minutes | Less than 5 seconds |
per patient | Â | Â |
