Table 2 Molecular subtyping error and accuracy in two test settings. Confusion matrices between true labels (RNA-seq-based IMS in columns) and predicted labels (WSI-based IMS in rows) at the patient-level for unselected (left) and low-confidence (right) by RNA-seq-based classification
Unselected test patients (N = 222) | Low-confidence test patients (N = 101) | |||||||
|---|---|---|---|---|---|---|---|---|
Basal-like | HER2-enriched | Luminal A | Luminal B | Basal-like | HER2-enriched | Luminal A | Luminal B | |
Basal-like | 13 | 0.9 | 0.9 | 0 | 0.96 | 0 | 0 | 0 |
HER2-enriched | 4.48 | 1.79 | 2.24 | 0.45 | 0.96 | 0 | 0 | 0 |
Luminal A | 1.35 | 1.79 | 46.19 | 2.69 | 4.81 | 1.92 | 43.27 | 1.92 |
Luminal B | 6.73 | 0.9 | 11.66 | 4.93 | 6.73 | 2.88 | 24.04 | 12.5 |
Patient-level accuracy 65.92% | Patient-level accuracy 56.73% | |||||||