Table 1 Summary of datasets

From: Mining multi-center heterogeneous medical data with distributed synthetic learning

Cardiac CTA

WHS64,65,66

ASOCA67,68

CAT0869

–

Train Subjects (Image#)

20 (3031)

32 (4642)

26 (3568)

–

Test Subjects (Image#)

40 (6360)

8 (1180)

6 (756)

–

Avg Spacing (mm3)

0.442 × 0.6

0.42 × 0.625

0.322 × 0.4

–

Scanner

Philips

Unknown

Siemens

–

BraTS1872,73,74

CBICA

TCIA

OTHER

–

Train Subjects (Image#)

69 (4638)

85 (5736)

14 (975)

–

Test Subjects (Image#)

19 (1165)

17 (1172)

6 (393)

–

Nuclei75 (tasks#)

Liver (t1)

Breast (t2)

Kidney (t3)

Prostate (t4)

Train Subjects (Nuclei#)

4 (1906)

4 (1508)

4 (4866)

4 (1634)

Test Subjects (Nuclei#)

2 (838)

2 (707)

2 (716)

2 (766)

  1. The cardiac CTA data were collected from three different sources and images were acquired from different devices with various spacings. The BraTS18 data was collected from different data centers with four modalities. In the missing modality completion setting, the modalities of different centers are misaligned. In the Nuclei dataset, each organ has a different data size and we conduct continual learning by using them sequentially.
Back to article page