Table 1. Characteristics of the Whole-Slide Images and Glass Slides in the Data Sets Used in the CAMELYON16 Challenge.
Data Set (N = 399 Slides and Images)a |
Hospital Providing the Slides and Images | Primary Tumor Histotypeb | Slides Containing Metastases, No. | No. of Lesions per Slide or Image, Median (Range) |
Total Slides or Images | |||
---|---|---|---|---|---|---|---|---|
IDC | Non-IDC | None | Macro | Micro | ||||
Training (n = 270 images) |
RUMC | 54 | 16 | 100 | 35 | 35 | 2 (1-20) | 170 |
UMCU | 30 | 10 | 60 | 26 | 14 | 3 (1-27) | 100 | |
Test (n=129 slides and images) |
RUMC | 23 | 6 | 50 | 14 | 15 | 2 (1-14) | 79 |
UMCU | 15 | 5 | 30 | 8 | 12 | 3 (1-25) | 50 |
Abbreviations: CAMELYON16, Cancer Metastases in Lymph Nodes Challenge 2016; IDC, infiltrating ductal carcinoma; RUMC, Radboud University Medical Center; UMCU, University Medical Center Utrecht.
All analyses in the training set were determined with whole-slide images. Analyses in the test were determined with whole-slide images by the algorithms and with glass slides by the panel of 11 pathologists (because diagnosing is most commonly done using a microscope in pathology labs).
Primary tumor histotypes included IDC and other histotypes (non-IDC).