Table 2.

Performance of the lung graph–based machine learning models in the identification of f-ILD on the testing set

Evaluation level	Method	AUC	Accuracy	Sensitivity	Specificity	PPV	NPV
Scan-level	Split 1	0.983	0.918	0.9	0.939	0.947	0.886
	Split 2	0.996	0.984	0.973	1	1	0.963
	Split 3	1	1	1	1	1	1
	Split 4	0.965	0.908	0.897	0.923	0.946	0.857
	Split 5	0.913	0.841	0.941	0.743	0.78	0.929
	Mean	0.971 ± 0.032	0.930 ± 0.057	0.942 ± 0.040	0.921 ± 0.094	0.935 ± 0.081	0.927 ± 0.051
Patient-level	Split 1	0.969	0.881	0.84	0.941	0.955	0.8
	Split 2	0.99	0.976	0.958	1	1	0.944
	Split 3	1	1	1	1	1	1
	Split 4	0.949	0.854	0.833	0.882	0.909	0.789
	Split 5	0.958	0.878	0.917	0.824	0.88	0.875
	Mean	0.973 ± 0.019	0.918 ± 0.059	0.910 ± 0.065	0.929 ± 0.068	0.949 ± 0.048	0.882 ± 0.081

All results are shown as mean values and standard deviations over the five random splits. Evaluation results (except AUC) of the proposed method were calculated by using the standard classification decision threshold of 0.5

AUC area under the curve, PPV positive predict value, NPV negative predict value