TF 2.0 is out! Get hands-on practice at TF World, Oct 28-31. Use code TF20 for 20% off select passes. Register now

curated_breast_imaging_ddsm

The CBIS-DDSM (Curated Breast Imaging Subset of DDSM) is an updated and standardized version of the Digital Database for Screening Mammography (DDSM). The DDSM is a database of 2,620 scanned film mammography studies. It contains normal, benign, and malignant cases with verified pathology information.

The default config is made of patches extracted from the original mammograms, following the description from http://arxiv.org/abs/1708.09427, in order to frame the task to solve in a traditional image classification setting.

Because special software and libraries are needed to download and read the images contained in the dataset, TFDS assumes that the user has downloaded the original DCIM files and converted them to PNG.

The following commands (or equivalent) should be used to generate the PNG files, in order to guarantee reproducible results:

find $DATASET_DCIM_DIR -name '*.dcm' |
xargs -n1 -P8 -I{} bash -c 'f={}; dcmj2pnm $f | convert - ${f/.dcm/.png}'

curated_breast_imaging_ddsm is configured with tfds.image.cbis_ddsm.CuratedBreastImagingDDSMConfig and has the following configurations predefined (defaults to the first one):

  • patches (v0.2.0) (Size: 2.01 MiB): Patches containing both calsification and mass cases, plus pathces with no abnormalities. Designed as a traditional 5-class classification task.

  • original-calc (v0.1.0) (Size: 1.06 MiB): Original images of the calcification cases compressed in lossless PNG.

  • original-mass (v0.1.0) (Size: 966.57 KiB): Original images of the mass cases compressed in lossless PNG.

curated_breast_imaging_ddsm/patches

Patches containing both calsification and mass cases, plus pathces with no abnormalities. Designed as a traditional 5-class classification task.

Versions:

  • 0.2.0 (default):
  • 2.0.0: New split API (https://tensorflow.org/datasets/splits)

Statistics

Split Examples
ALL 65,130
TRAIN 49,780
TEST 9,770
VALIDATION 5,580

Features

FeaturesDict({
    'id': Text(shape=(), dtype=tf.string),
    'image': Image(shape=(None, None, 1), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=5),
})

Urls

curated_breast_imaging_ddsm/original-calc

Original images of the calcification cases compressed in lossless PNG.

Versions:

  • 0.1.0 (default):
  • 2.0.0: New split API (https://tensorflow.org/datasets/splits)

Statistics

Split Examples
ALL 1,511
TRAIN 1,227
TEST 284

Features

FeaturesDict({
    'abnormalities': Sequence({
        'assessment': ClassLabel(shape=(), dtype=tf.int64, num_classes=6),
        'calc_distribution': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
        'calc_type': ClassLabel(shape=(), dtype=tf.int64, num_classes=48),
        'id': Tensor(shape=(), dtype=tf.int32),
        'mask': Image(shape=(None, None, 1), dtype=tf.uint8),
        'pathology': ClassLabel(shape=(), dtype=tf.int64, num_classes=3),
        'subtlety': ClassLabel(shape=(), dtype=tf.int64, num_classes=6),
    }),
    'breast': ClassLabel(shape=(), dtype=tf.int64, num_classes=2),
    'id': Text(shape=(), dtype=tf.string),
    'image': Image(shape=(None, None, 1), dtype=tf.uint8),
    'patient': Text(shape=(), dtype=tf.string),
    'view': ClassLabel(shape=(), dtype=tf.int64, num_classes=2),
})

Urls

curated_breast_imaging_ddsm/original-mass

Original images of the mass cases compressed in lossless PNG.

Versions:

  • 0.1.0 (default):
  • 2.0.0: New split API (https://tensorflow.org/datasets/splits)

Statistics

Split Examples
ALL 1,514
TRAIN 1,166
TEST 348

Features

FeaturesDict({
    'abnormalities': Sequence({
        'assessment': ClassLabel(shape=(), dtype=tf.int64, num_classes=6),
        'id': Tensor(shape=(), dtype=tf.int32),
        'mask': Image(shape=(None, None, 1), dtype=tf.uint8),
        'mass_margins': ClassLabel(shape=(), dtype=tf.int64, num_classes=20),
        'mass_shape': ClassLabel(shape=(), dtype=tf.int64, num_classes=21),
        'pathology': ClassLabel(shape=(), dtype=tf.int64, num_classes=3),
        'subtlety': ClassLabel(shape=(), dtype=tf.int64, num_classes=6),
    }),
    'breast': ClassLabel(shape=(), dtype=tf.int64, num_classes=2),
    'id': Text(shape=(), dtype=tf.string),
    'image': Image(shape=(None, None, 1), dtype=tf.uint8),
    'patient': Text(shape=(), dtype=tf.string),
    'view': ClassLabel(shape=(), dtype=tf.int64, num_classes=2),
})

Urls

Citation

@misc{CBIS_DDSM_Citation,
  doi = {10.7937/k9/tcia.2016.7o02s9cy},
  url = {https://wiki.cancerimagingarchive.net/x/lZNXAQ},
  author = {Sawyer-Lee,  Rebecca and Gimenez,  Francisco and Hoogi,  Assaf and Rubin,  Daniel},
  title = {Curated Breast Imaging Subset of DDSM},
  publisher = {The Cancer Imaging Archive},
  year = {2016},
}
@article{TCIA_Citation,
  author = {
    K. Clark and B. Vendt and K. Smith and J. Freymann and J. Kirby and
    P. Koppel and S. Moore and S. Phillips and D. Maffitt and M. Pringle and
    L. Tarbox and F. Prior
  },
  title = { {The Cancer Imaging Archive (TCIA): Maintaining and Operating a
  Public Information Repository}},
  journal = {Journal of Digital Imaging},
  volume = {26},
  month = {December},
  year = {2013},
  pages = {1045-1057},
}
@article{DBLP:journals/corr/abs-1708-09427,
  author    = {Li Shen},
  title     = {End-to-end Training for Whole Image Breast Cancer Diagnosis using
               An All Convolutional Design},
  journal   = {CoRR},
  volume    = {abs/1708.09427},
  year      = {2017},
  url       = {http://arxiv.org/abs/1708.09427},
  archivePrefix = {arXiv},
  eprint    = {1708.09427},
  timestamp = {Mon, 13 Aug 2018 16:48:35 +0200},
  biburl    = {https://dblp.org/rec/bib/journals/corr/abs-1708-09427},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}