Table 5.

Performance of model versus radiologists of clinical experience <5 years, 5–10 years, and >10 years in diagnosing malignant nodules on the test set and its subsets.

	Diagnostic method	AUC	Accuracy	Sensitivity	Specificity	Precision
Test set	Model	0.924	0.851	0.881	0.839	0.673
	Radiologist <5 yr	0.818	0.868	0.707	0.928	0.784
	Radiologist 5–10 yr	0.843	0.864	0.798	0.888	0.726
	Radiologist >10 yr	0.848	0.858	0.826	0.87	0.701
	P-Value*	<0.01	0.781	<0.01	<0.01	0.016
	P-Value**	<0.01	1.000	0.001	0.04	0.346
	P-Value***	<0.01	0.733	0.777	0.3	0.752
HT subset	Model	0.924	0.852	0.881	0.846	0.540
	Radiologist <5 yr	0.824	0.897	0.723	0.924	0.588
	Radiologist 5–10 yr	0.857	0.875	0.831	0.882	0.514
	Radiologist >10 yr	0.863	0.863	0.862	0.863	0.487
	P-Value*	<0.01	0.401	0.001	0.003	0.226
	P-Value**	<0.01	0.928	0.060	0.312	0.811
	P-Value***	<0.01	0.787	0.486	1.000	0.874
Normal subset	Model	0.906	0.843	0.871	0.822	0.784
	Radiologist <5 yr	0.825	0.842	0.712	0.938	0.893
	Radiologist 5–10 yr	0.846	0.853	0.797	0.894	0.847
	Radiologist >10 yr	0.844	0.85	0.804	0.885	0.837
	P-Value*	<0.01	0.603	0	0	0.017
	P-Value**	<0.01	0.916	0.01	0.035	0.179
	P-Value***	<0.01	0.833	0.015	0.072	0.272

P-Value^* is that of model versus radiologist with <5 years’ clinical experience; P-Value^** is that of model versus radiologist with 5–10 years’ clinical experience; P-Value^*** is that of model versus radiologist with >10 years’ clinical experience; AUC, Areas under the ROC curve.

All metrics were the average of 10-folds.