. Author manuscript; available in PMC: 2023 Aug 1.

Published in final edited form as: IEEE Trans Med Imaging. 2022 Aug 1;41(8):1990–2003. doi: 10.1109/TMI.2022.3153322

TABLE V:

The AUROC and AUPRC results of various models to classify the 13 classes (12 diseases and Support Devices) on MIMIC-CXR dataset. The upper part is the AUROC results and the lower part is the AUPRC results.

		EC	Card	AO	LL	Edem	Cons	Pneu1	Atel	Pneu2	PE	PO	Frac	SD	mean

AUROC	A-GCN-PPS	0.756	0.826	0.749	0.791	0.905	0.817	0.725	0.817	0.883	0.919	0.826	0.725	0.897	0.818
	A-GCN-APS	0.752	0.816	0.743	0.768	0.887	0.801	0.723	0.804	0.877	0.901	0.826	0.764	0.879	0.811
	AlexNet	0.745	0.819	0.738	0.721	0.901	0.805	0.703	0.812	0.850	0.911	0.791	0.666	0.889	0.796

	R-GCN-PPS	0.764	0.828	0.760	0.828	0.909	0.821	0.764	0.826	0.895	0.926	0.849	0.761	0.907	0.834
	R-GCN-APS	0.760	0.820	0.761	0.778	0.892	0.818	0.744	0.817	0.897	0.909	0.853	0.825	0.890	0.828
	ResNet50	0.752	0.822	0.756	0.736	0.905	0.816	0.729	0.822	0.862	0.923	0.817	0.668	0.900	0.808

	V-GCN-PPS	0.767	0.826	0.760	0.826	0.908	0.826	0.761	0.824	0.897	0.926	0.844	0.761	0.908	0.833
	V-GCN-APS	0.749	0.818	0.762	0.785	0.889	0.812	0.749	0.819	0.894	0.908	0.858	0.820	0.891	0.827
	VGGNet16BN	0.743	0.818	0.748	0.715	0.904	0.816	0.717	0.821	0.856	0.920	0.796	0.678	0.900	0.802

	p-val (RM-ANOVA)	^**	^***	^*	^***	^***	^**	^***	^***	^***	^***	^***	^***	^***	^***
	PPS >APS	^*	^***		^**	^***	^*	^*	^**		^***			^***	^***
	PPS >base	^**	^***	^**	^***	^***	^**	^**	^***	^***	^**	^***	^***	^***	^***

AUPRC	A-GCN-PPS	0.096	0.478	0.407	0.150	0.528	0.154	0.203	0.455	0.313	0.738	0.068	0.073	0.698	0.335
	A-GCN-APS	0.094	0.461	0.398	0.122	0.480	0.145	0.191	0.436	0.313	0.690	0.087	0.099	0.659	0.321
	AlexNet	0.087	0.465	0.397	0.085	0.510	0.138	0.182	0.456	0.228	0.726	0.044	0.041	0.687	0.311

	R-GCN-PPS	0.102	0.479	0.420	0.224	0.548	0.168	0.233	0.476	0.370	0.760	0.081	0.102	0.730	0.361
	R-GCN-APS	0.094	0.468	0.413	0.147	0.490	0.145	0.245	0.452	0.403	0.707	0.127	0.203	0.679	0.352
	ResNet50	0.092	0.473	0.417	0.100	0.535	0.155	0.213	0.472	0.288	0.755	0.062	0.044	0.713	0.332

	V-GCN-PPS	0.101	0.475	0.421	0.191	0.539	0.166	0.230	0.469	0.368	0.753	0.091	0.109	0.726	0.357
	V-GCN-APS	0.098	0.470	0.419	0.149	0.489	0.156	0.233	0.458	0.376	0.703	0.118	0.178	0.679	0.348
	VGGNet16BN	0.088	0.468	0.409	0.082	0.520	0.152	0.195	0.470	0.246	0.744	0.052	0.046	0.714	0.322

	p-val (RM-ANOVA)	^***	^**		^***	^***	^**	^**	^***	^***	^***	^***	^**	^***	^***
	PPS >APS	^*	^**	^**	^**	^***	^**		^**		^***			^***	^**
	PPS >base	^***	^**	^*	^**	^***	^***	^**		^***	^**	^**	^**	^**	^***

*, ** and *** denote the significance levels of 0.1, 0.05 and 0.01, respectively. For each disease, the best results are bolded. The red text means our ImageGCN can perform better than the corresponding two baseline models.

EC: Enlarged Cardiomediastinum; Card: Cardiomegaly; AO: Airspace Opacity; LL: Lung Lesion; Edem: Edema; Cons: Consolidation; Pneu1: Pneumonia; Atel: Atelectasis; Pneu2: Pneumothorax; PE: Pleural Effusion; PO: Pleural Other; Frac: Fracture; SD: Support Devices