Table 5 Comparison of 3 AI models from 3 cross-validation splits in 1STL algorithm against gynecological pathologists’ diagnoses.
From: Deep learning-based histotype diagnosis of ovarian carcinoma whole-slide pathology images
Case | AI predicted diagnosis | Reference diagnosis | Independent study review diagnosis | Cancer Registry diagnosis | COSPv3 histotype prediction | Immunophenotype | |||
|---|---|---|---|---|---|---|---|---|---|
Model 1 | Model 2 | Model 3 | Overall Majority of models | ||||||
A | HGSC | HGSC | HGSC | HGSC | ENOC | HGSC with transitional differentiation | Endometrioid | ENOC | WT1-;p53wt; PR− |
B | HGSC | HGSC | ENOC | HGSC | ENOC | HGSC | NOS carcinoma | HGSC | WT1-;p53abn;PR− |
C | MUC | MUC | CCOC | MUC | ENOC | CCOC; IHC needed for diagnosis | Endometrioid | ENOC | WT1-;p53wt;PR-;NapsinA− |
D | ENOC | ENOC | ENOC | ENOC | HGSC | HGSC with transitional differentiation | Serous | HGSC | WT1+;p53abn;PR+ |
E | HGSC | HGSC | ENOC | HGSC | ENOC | ENOC | Endometrioid | ENOC | WT1-;p53wt; PR+ |
F | CCOC | CCOC | CCOC | CCOC | HGSC | CCOC; IHC needed for diagnosis | Clear cell | HGSC | WT1-;p53abn;PR-;NapsinA− |
G | ENOC | ENOC | ENOC | ENOC | MUC | ENOC | Mucinous | MUC | PR− |
H | MUC | MUC | MUC | MUC | ENOC | ENOC | Endometrioid | ENOC | PR− |
