CARDIAG: A Dense Segment Classification Benchmark of Deep Learning Architectures for Coronary Angiography

This record contains CARDIAG: a benchmark for coronary artery SYNTAX segmentation and plenty other tasks (`CARDIAG.zip`). It consists of 644 data points, each of which provided with:

• original X-ray
• SYNTAX segments
• expert uncertainty masks
• catheter
• stenoses
• supplementary keyframes
• acquisition metadata including:
  • primary angle
  • secondary angle
  • SID
  • SOD
  • imager pixel spacing (horizontal and vertical)
• extra metadata:
  • medical center name (hashed)
  • sample ID

For more information on the acquisition of the data consult the attached study. We provide pretrained weights of the models discussed in the supplementary study in `models.zip` archive. We provide source code for processing the data and training models: https://github.com/cvlab-ai/cardiag-benchmark

NOTE: some medical centers have different names in data, therefore below we attach a mapping hash -> center ID:

Site 1

101b1418aebdcce57689ef5a5e9d1781a1bb0275c7ef5244852d04b1a61beea2

299532f16d45f9e04a31b311a39d620cf4a958b8b56c51a6ca683ce21dced708

Site 2

46068363100ef9a99ab1d4af987c2a0bf9dcdee0f4a44596a17a8888ce4f6fc9

d04e8912730bd49fd176710ff630888e7f70906cd31f0f16a63e54d7c8076f7d

a9f27c29b86cfa3e756b040a44bc2108be1b5cac48f4c2232b2f6f51ba46379d

Site 3

938b824173e8cceef065c6338dfc9e7512abbcb1580192cb6b1a73787af4fc30

Site 4

77c5079fe210d5e0649500e3c844c4777f5c193c1397f0aab7b887fe4abf576a

7c38f06306e44c0cc87e6368fc0d77840193311ef8cd9e0694cf7ec5d924a1ed

Site 5

8caa248386239d9c3733fbe46302767c36ffb38628ae94c9f583e12c92eb7b8b