Table 3.
Results obtained performing MIL and the barcode analysis on single-scale and multi-scale inputs.
| MIL | Barcode | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| 5x | 10x | 20x | 5x + 10x | 5x + 10x + 20x | 5x | 10x | 20x | 5x + 10x | 5x + 10x + 20x | |
| Weighted F1 | 0.88 | 0.86 | 0.91 | 0.93 | 0.94 | 0.69 | 0.58 | 0.65 | 0.72 | 0.74 |
| Macro F1 | 0.87 | 0.85 | 0.88 | 0.90 | 0.94 | 0.67 | 0.55 | 0.62 | 0.70 | 0.73 |
| MCC | 0.88 | 0.86 | 0.9 | 0.93 | 0.94 | 0.68 | 0.57 | 0.64 | 0.71 | 0.74 |
| ACC | 0.86 | 0.86 | 0.95 | 0.95 | 0.95 | 0.53 | 0.48 | 0.58 | 0.59 | 0.60 |
| BLCA | 0.92 | 0.90 | 0.95 | 0.97 | 0.98 | 0.60 | 0.20 | 0.45 | 0.49 | 0.55 |
| BRCA | 0.84 | 0.86 | 0.89 | 0.95 | 0.84 | 0.62 | 0.49 | 0.59 | 0.68 | 0.71 |
| CESC | 0.83 | 0.86 | 0.88 | 0.90 | 0.95 | 0.62 | 0.68 | 0.57 | 0.71 | 0.73 |
| CHOL | 0.83 | 0.80 | 0.71 | 0.86 | 0.83 | 0.33 | 0.36 | 0.36 | 0.36 | 0.36 |
| COAD | 0.84 | 0.82 | 0.88 | 0.95 | 0.93 | 0.68 | 0.45 | 0.56 | 0.65 | 0.76 |
| DLBC | 0.50 | 0.00 | 0.00 | 0.00 | 1.00 | 0.67 | 0.00 | 0.00 | 0.67 | 0.67 |
| ESCA | 0.95 | 0.76 | 0.93 | 0.95 | 0.95 | 0.68 | 0.26 | 0.62 | 0.65 | 0.67 |
| GBM | 0.89 | 0.90 | 0.84 | 0.95 | 0.87 | 0.82 | 0.79 | 0.71 | 0.86 | 0.86 |
| HNSC | 0.78 | 0.89 | 0.93 | 0.90 | 0.90 | 0.70 | 0.34 | 0.65 | 0.76 | 0.79 |
| KICH | 0.97 | 0.97 | 1.00 | 0.97 | 0.98 | 0.90 | 0.82 | 0.75 | 0.87 | 0.85 |
| KIRC | 0.95 | 0.97 | 0.90 | 0.95 | 0.97 | 0.87 | 0.90 | 0.90 | 0.92 | 0.95 |
| KIRP | 0.88 | 0.95 | 0.92 | 0.93 | 1.00 | 0.72 | 0.47 | 0.55 | 0.78 | 0.67 |
| LGG | 0.92 | 0.90 | 0.93 | 0.90 | 0.90 | 0.95 | 0.90 | 0.90 | 0.95 | 0.95 |
| LIHC | 0.92 | 0.84 | 0.89 | 0.92 | 0.95 | 0.55 | 0.58 | 0.57 | 0.59 | 0.63 |
| LUAD | 0.79 | 0.76 | 0.77 | 0.85 | 0.84 | 0.60 | 0.58 | 0.64 | 0.63 | 0.70 |
| LUSC | 0.72 | 0.68 | 0.68 | 0.73 | 0.79 | 0.35 | 0.28 | 0.47 | 0.38 | 0.49 |
| MESO | 0.83 | 0.85 | 0.97 | 1.00 | 1.00 | 0.37 | 0.21 | 0.21 | 0.47 | 0.32 |
| OV | 0.90 | 0.98 | 1.00 | 0.98 | 0.95 | 0.68 | 0.59 | 0.53 | 0.74 | 0.78 |
| PAAD | 0.90 | 0.90 | 0.92 | 0.95 | 0.98 | 0.83 | 0.51 | 0.80 | 0.79 | 0.82 |
| PCPG | 0.97 | 0.98 | 1.00 | 1.00 | 0.97 | 0.76 | 0.68 | 0.88 | 0.81 | 0.90 |
| PRAD | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 0.97 | 0.87 | 0.87 | 0.97 | 0.97 |
| SARC | 0.89 | 0.90 | 0.97 | 0.92 | 0.95 | 0.61 | 0.70 | 0.58 | 0.73 | 0.76 |
| SKCM | 0.78 | 0.92 | 0.87 | 0.92 | 0.92 | 0.55 | 0.59 | 0.60 | 0.56 | 0.74 |
| STAD | 0.92 | 0.82 | 0.76 | 0.87 | 0.89 | 0.57 | 0.21 | 0.40 | 0.53 | 0.53 |
| TGCT | 0.92 | 0.95 | 0.98 | 1.00 | 1.00 | 0.86 | 0.74 | 0.80 | 0.89 | 0.85 |
| THCA | 0.98 | 0.95 | 1.00 | 0.98 | 1.00 | 0.95 | 0.88 | 0.90 | 0.95 | 0.95 |
| THYM | 0.93 | 0.90 | 1.00 | 0.95 | 1.00 | 0.83 | 0.85 | 0.87 | 0.85 | 0.85 |
| UCEC | 0.86 | 0.78 | 0.92 | 0.95 | 0.91 | 0.76 | 0.65 | 0.67 | 0.71 | 0.74 |
| UCS | 0.80 | 0.70 | 0.83 | 0.84 | 0.90 | 0.25 | 0.54 | 0.52 | 0.60 | 0.63 |
The results are obtained by training the models on a smaller subsample of the dataset (2,727 slides) and testing on a 20% split. The value of the best performing magnification (or combination thereof) is highlighted in green. The top 3 rows show the aggregated classification results: weighted F1, macro F1 and MCC scores for the overall classification across all different tumor types on the test set.