Table 3.

Superpatches evaluation using polyserial correlation coefficient.

Model Name	ACC (147)	BLCA (64)	BRCA (348)	CESC (61)	COAD (65)	ESCA (312)	HNSC (324)	KIRC (319)
Baseline	–	0.720	0.552	0.679	0.329	–	–	–
VGG-16	0.879	0.787	0.745	0.592	0.688	0.777	0.904	0.515
ResNet-34	0.925	0.740	0.797	0.654	0.658	0.810	0.883	0.599
Incep-V4	0.963	0.744	0.797	0.667	0.695	0.805	0.897	0.598
Model Name	LIHC (248)	LUAD (63)	LUSC (65)	MESO (271)	OV (158)	PAAD (440)	PRAD (66)	READ (62)
Baseline	–	0.615	0.658	–	–	0.695	0.819	0.706
VGG-16	0.891	0.670	0.830	0.840	0.565	0.886	0.885	0.702
ResNet-34	0.872	0.733	0.775	0.805	0.527	0.874	0.862	0.715
Incep-V4	0.854	0.617	0.789	0.818	0.635	0.870	0.818	0.811
Model Name	SARC (299)	SKCM (67)	STAD (63)	TGCT (303)	THYM (324)	UCEC (64)	UVM (64)	ALL (4198)
Baseline	–	0.666	0.728	–	–	0.692	0.681	–
VGG-16	0.912	0.816	0.713	0.859	0.774	0.667	0.896	0.807 (0.77 ± 0.12)
ResNet-34	0.932	0.794	0.821	0.799	0.765	0.766	0.899	0.808 (0.78 ± 0.10)
Incep-34	0.921	0.822	0.752	0.823	0.790	0.742	0.913	0.820 (0.79 ± 0.10)

The number in brackets indicated the number of superpatches in the respective cancer type. Baseline is the model developed in (33).

Highest polyserial correlation in each dataset (cancer type) is indicated in bold.