Validating the accuracy of deep learning for the diagnosis of pneumonia on chest x-ray against a robust multimodal reference diagnosis: a post hoc analysis of two prospective studies

European Radiology Experimental

Table 3 ROC-AUC and statistical comparisons of diagnostic performances (PneumOld-CT cohort)

		Clinicians	Radiologist 1	Radiologist 2	AI	LDCT
	ROC-AUC	0.577 (0.509–0.646)	0.782 (0.726–0.839)	0.726 (0.662–0.791)	0.738 (0.664–0.812)	0.820 (0.769–0.871)
Clinicians	0.577 (0.509–0.646)
Radiol. 1	0.782 (0.726–0.839)	p < 0.001
Radiol. 2	0.726 (0.662–0.791)	p < 0.001	p = 0.134
AI	0.738 (0.664–0.812)	p < 0.001	p = 0.269	p = 0.768
LDCT	0.820 (0.769–0.871)	p < 0.001	p = 0.233	p = 0.013	p = 0.065

ROC-AUC results are reported in percent along with their 95% confidence interval. Statistical comparison of two sets of predictions by area under the receiver operating characteristic curve using method from Sun and Xu. [reference #[31]. AI, Artificial intelligence; CT, Computed tomography; LDCT, Low-dose computed tomography; ROC-AUC, Area under the receiver operating characteristic curve