Table 1. Reconstruction accuracy in training and validation datasets for CTransPath encoders.

We compare reconstruction accuracy from the real and reconstructed images for HistoXGAN and other architectures for embedding images in GAN latent space. For comparison, we use encoders designed to recreate images from a StyleGAN2 model trained identically to the HistoXGAN model. The LPIPS/DISTS encoder uses an equal ratio of LPIPS/DISTS loss between the real and reconstructed images to train the encoder. The Single Layer and Encoder4Editing encoders are trained to minimize L1 loss between CTransPath feature vector of the real and reconstructed images.

Source		n	n tiles	LPIP/DISTSL1 loss	Single layer L1 loss	Encoder4Editing L1 loss	HistoXGAN L1 loss
TCGA	ACC	56	78789	0.086 (0.010)	0.054 (0.010)	0.045 (0.006)	0.035 (0.004)
TCGA	BLCA	378	342,463	0.082 (0.009)	0.054 (0.015)	0.044 (0.006)	0.034 (0.004)
TCGA	BRCA	943	454,985	0.087 (0.010)	0.055 (0.017)	0.044 (0.006)	0.033 (0.004)
TCGA	CESC	267	165,088	0.083 (0.010)	0.063 (0.021)	0.044 (0.006)	0.034 (0.004)
TCGA	CHOL	38	51,414	0.087 (0.009)	0.057 (0.019)	0.044 (0.006)	0.034 (0.004)
TCGA	COADREAD	428	195,493	0.092 (0.013)	0.071 (0.023)	0.046 (0.007)	0.035 (0.004)
TCGA	DLBC	43	32,073	0.076 (0.011)	0.063 (0.022)	0.045 (0.010)	0.032 (0.005)
TCGA	ESCA	147	99,625	0.085 (0.010)	0.057 (0.017)	0.045 (0.007)	0.035 (0.005)
TCGA	HNSC	401	166,061	0.083 (0.009)	0.060 (0.019)	0.045 (0.006)	0.035 (0.004)
TCGA	KICH	101	105,461	0.094 (0.010)	0.056 (0.009)	0.048 (0.006)	0.036 (0.005)
TCGA	KIRP	270	234,740	0.091 (0.013)	0.070 (0.022)	0.049 (0.007)	0.036 (0.005)
TCGA	LGG	464	155,579	0.081 (0.009)	0.053 (0.014)	0.044 (0.006)	0.031 (0.005)
TCGA	LIHC	359	331,769	0.090 (0.013)	0.058 (0.019)	0.045 (0.006)	0.033 (0.004)
TCGA	LUAD	467	335,499	0.088 (0.010)	0.062 (0.019)	0.046 (0.007)	0.035 (0.004)
TCGA	LUSC	474	370,542	0.083 (0.009)	0.061 (0.019)	0.045 (0.006)	0.035 (0.004)
TCGA	MESO	73	42,242	0.083 (0.009)	0.053 (0.011)	0.044 (0.006)	0.034 (0.004)
TCGA	OV	104	108,620	0.085 (0.010)	0.061 (0.019)	0.045 (0.007)	0.035 (0.005)
TCGA	PAAD	168	119,144	0.085 (0.009)	0.053 (0.012)	0.046 (0.007)	0.035 (0.005)
TCGA	PCPG	173	207,803	0.083 (0.009)	0.066 (0.022)	0.044 (0.006)	0.035 (0.004)
TCGA	PRAD	394	202,439	0.091 (0.010)	0.057 (0.015)	0.046 (0.007)	0.034 (0.004)
TCGA	SARC	250	326,262	0.084 (0.011)	0.059 (0.019)	0.047 (0.007)	0.033 (0.005)
TCGA	SKCM	418	322,527	0.080 (0.009)	0.063 (0.023)	0.044 (0.006)	0.034 (0.004)
TCGA	STAD	371	283,734	0.086 (0.011)	0.055 (0.014)	0.046 (0.008)	0.035 (0.005)
TCGA	TGCT	129	109,128	0.082 (0.008)	0.057 (0.015)	0.045 (0.007)	0.034 (0.005)
TCGA	THCA	480	279,362	0.091 (0.010)	0.065 (0.022)	0.047 (0.007)	0.035 (0.005)
TCGA	THYM	114	157,626	0.078 (0.010)	0.080 (0.028)	0.045 (0.008)	0.033 (0.006)
TCGA	UCEC	477	351,784	0.084 (0.010)	0.071 (0.025)	0.045 (0.006)	0.035 (0.004)
TCGA	UCS	53	67,853	0.082 (0.009)	0.061 (0.019)	0.045 (0.006)	0.035 (0.005)
TCGA	UVM	80	35,766	0.075 (0.009)	0.063 (0.020)	0.045 (0.007)	0.033 (0.005)
CPTAC	BRCA	105	56,318	0.084 (0.008)	0.067 (0.016)	0.053 (0.008)	0.040 (0.005)
CPTAC	COADREAD	104	76,915	0.086 (0.009)	0.071 (0.015)	0.056 (0.007)	0.041 (0.005)
CPTAC	GBM	177	143,858	0.080 (0.010)	0.062 (0.016)	0.048 (0.007)	0.038 (0.006)
CPTAC	HNSC	108	35,064	0.080 (0.009)	0.062 (0.018)	0.049 (0.007)	0.038 (0.005)
CPTAC	LUAD	221	425,473	0.087 (0.009)	0.057 (0.012)	0.048 (0.006)	0.038 (0.004)
CPTAC	LUSC	202	359,394	0.081 (0.008)	0.066 (0.021)	0.046 (0.006)	0.036 (0.004)
CPTAC	PAAD	164	99,093	0.082 (0.009)	0.059 (0.016)	0.049 (0.008)	0.037 (0.006)
CPTAC	UCEC	247	184,767	0.083 (0.010)	0.074 (0.021)	0.049 (0.008)	0.038 (0.006)