PAM50 Subtype Classification on TCGA-BRCA
86.95AccuracyNSGA-II-based patch selection + a custom CNN classifier
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| NSGA-II-based patch selection + a custom CNN classifierInternal dataset for training=TCGA-BRCA, Dataset modality=Histopathology (WSIs)2026.04 | 86.95 | 88 | — | 0.0098 | 89.95 | 77 | — | |
| Transfer Learning (CTL)+ ResNet50Internal dataset for training=TCGA-BRCA (split into TCGA-A2, TCGA-BH, TCGA-E2, and held-out TCGA-EW), Dataset modality=WSIs2026.04 | 81.8 | — | 72 | — | — | — | — | |
| InceptionV3Internal dataset for training=TCGA-BRCA, Dataset modality=WSIs2026.04 | 79 | — | — | — | — | — | — | |
| a multi-instance learning based framework (DPMIL)Internal dataset for training=TCGA-BRCA, Dataset modality=WSIs2026.04 | 74.85 | 69 | — | — | — | — | — | |
| CNNs+ SVMInternal dataset for training=TCGA-BRCA, Dataset modality=WSIs2026.04 | 65.92 | — | — | — | — | — | — | |
| Multimodal deep learning with feature extractionInternal dataset for training=TCGA-BRCA, Dataset modality=Gene expression, CNV data, and WSIs2026.04 | — | — | — | 0.8807 | — | — | — | |
| ResNet-18 + XGBoostInternal dataset for training=TCGA-BRCA, Dataset modality=Histopathology (WSIs)2026.04 | — | 72.7 | — | — | 74.1 | 72.6 | 93 |