Table 1 Dataset summary for slides and cases used in model development and evaluation.
From: Determining breast cancer biomarker status and associated morphological features using deep learning
Tertiary teaching hospital | Medical laboratory | Tertiary teaching hospital | Tertiary teaching hospital | TCGA(36 sites) | ||
---|---|---|---|---|---|---|
DLS Stage 1 (patch-level): uses paired H&E and IHC slides from custom sectioning protocol | NA | |||||
No. of cases(train / tune / test) | 70 / 30/ 64 | 70 / 30/ 0 | ||||
No. of H&E slides(Train / tune / test) | 205 / 80/ 181 | 206 / 85/ 0 | ||||
No. of patches | See Table 2 | |||||
DLS Stage 2 (slide-level): uses biomarker status from the original report | Train | Tune | Test | |||
No. of cases | 164 | 100 | 164** | 340 | 909 | |
No. of H&E slides | 466 | 291 | 1,377 | 2,313 | 961 | |
ER status* (pos / neg) | 103 / 47 | 91 / 7 | 103 / 47 | 280 / 58 | 679 / 191 | |
PR status* (pos / neg) | 93 / 56 | 85 / 13 | 93 / 56 | 251 / 87 | 573 / 283 | |
Her2 status* (pos / neg) | 16 / 34 | 4 / 81 | 16 / 34 | 11 / 78 | 131 / 739 | |
Nottingham grade 1 / 2 / 3 | 46 / 69/ 49 | 44 / 36/ 20 | 46 / 69/ 49 | 135 / 126/ 79 | 387 / 295 / 227 |