Table 5 .

Comparison of diagnosis accuracy with one source batch for training and another target batch for testing

Source	Target	Baseline	ComBat	Ratio_G	fSVA	ResNet	NormAE	Remove_R	Ours
1	2	0.753	0.778	0.798	0.773	0.791	0.805	0.852	0.889
1	3	0.813	0.797	0.858	0.836	0.803	0.812	0.856	0.879
2	1	0.799	0.817	0.821	0.857	0.824	0.827	0.839	0.875
2	3	0.828	0.851	0.818	0.829	0.852	0.866	0.833	0.870
3	1	0.876	0.861	0.864	0.854	0.868	0.889	0.863	0.907
3	2	0.763	0.759	0.754	0.824	0.805	0.799	0.821	0.884
Overall		0.805	0.811	0.819	0.829	0.824	0.833	0.844	0.884

“Baseline” denotes classification based on raw input data without any calibration for batch effect removal