Squrve-Benchmarks / spider2 /lite /external /dicom_dataset_selection.md
satissss's picture
Upload 174 files
47d665b verified

Detailed requirements include:

  • The slides must belong to the TCGA-LUAD or TCGA-LUSC collections;
  • The slides must be JPEG or JPEG2000 compressed;
  • The slides must be digital images and exclude non-volume ones;
  • The slides must contain either normal (17621005) or tumor (86049000) tissue, identified by specific DICOM codes;
  • The slides must be prepared using the "Tissue freezing medium" embedding method;

With respect to the output, it should contain the following basic metadata and attributes related to slide microscopy images. Concretely,

  • digital slide ID: unique numeric identifier of a digital slide, i.e., a digital image of a physical slide;
  • case ID: unique numeric identifier of the study in the context of which the ditial slide was created;
  • physical slide ID: unique numeric identifier of the physical slide as prepared in the wet lab;
  • patient ID: unique numeric identifier of the patient from whose tissue the physical slide was obtained;
  • collection ID: numeric or character sequence describing the dataset the physical slide is part of;
  • instance ID: universally unique identifier of the DICOM instance;
  • GCS URL: how to access the DICOM file;
  • width/height: image width and height in pixels, respectively;
  • pixel spacing: image pixel spacing in mm/px;
  • compression type (either value jpeg, jpeg2000, or other).

And the target two labels are:

  • tissue_type: either normal or tumor;
  • cancer_subtype: either luad or lscc. Sort the results according to instance ID in ascending order.