Table 2.

The diagnostic performance of TL and non-TL in GM and EM

Stage and method	AUC	p value	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)	LR+	LR−
Training cohort
S4
GM non-TL	0.957 (0.945–0.968)	< .001	91.0	88.4	88.8	89.9	7.2	0.1
TL	0.994 (0.945–0.968)		95.1	95.6	96.3	95.8	21.3	0.0
EM non-TL	0.991 (0.983–0.995)	.001	96.3	93.9	97.4	96.6	18.6	0.0
TL	1.0 (0.992–1.0)		100.0	100.0	100.0	100.0	/	0.0
≥ S3
GM non-TL	0.978 (0.966–0.989)	.002	92.1	93.1	95.3	88.5	13.3	0.1
TL	0.992 (0.981–0.997)		96.9	95.0	97.1	95.0	19.0	0.0
EM non-TL	0.985 (0.977–0.990)	.002	96.5	89.1	95.0	92.7	8.7	0.0
TL	0.993 (0.983–0.998)		97.4	95.8	98.1	97.0	23.4	0.0
≥ S2
GM non-TL	0.985 (0.976–0.991)	.001	93.9	95.2	97.58	87.27	19.5	0.1
TL	0.992 (0.9719–1.0)		95.7	94.4	97.50	95.04	22.3	0.0
EM non-TL	0.988 (0.981–0.991)	.001	92.8	96.9	98.33	84.77	20.9	0.1
TL	0.994 (0.981–0.998)		98.8	96.0	98.08	97.04	25.8	0.0
Test cohort
S4
GM non-TL	0.852 (0.785–0.901)	.002	81.6	75.3	79.8	77.5	3.3	0.2
TL	0.897 (0.831–0.940)		86.0	83.6	87.8	81.3	5.2	0.2
EM non-TL	0.862 (0.787–0.920)	.002	81.1	75.9	82.1	74.5	3.4	0.3
TL	0.921 (0.897–0.951)		89.1	87.0	92.8	81.3	6.9	0.1
≥ S3
GM non-TL	0.843 (0.768–0.901)	.002	80.5	66.7	78.5	72.3	2.4	0.3
TL	0.885 (0.827–0.919)		87.7	84.3	92.1	76.8	5.6	0.2
EM non-TL	0.861 (0.785–0.916)	.002	83.6	82.5	91.8	76.7	5.1	0.1
TL	0.910 (0.853–0.952)		85.2	90.0	93.1	73.3	8.1	0.2
≥ S2
GM non-TL	0.844 (0.769–0.902)	.001	70.0	74.1	76.2	67.2	2.7	0.4
TL	0.882 (0.825–0.918)		87.1	83.3	89.9	70.6	5.3	0.1
EM non-TL	0.867 (0.793–0.917)	.003	85.9	81.4	90.1	74.7	4.6	0.2
TL	0.907 (0.793–0.917)		87.9	83.2	90.2	76.0	5.5	0.2

Data in parentheses are 95% confidence intervals

Non-TL non-transfer learning, TL transfer learning, GM gray scale modality, EM elastogram modality, NPV negative predictive value, PPV positive predictive value, LR+ positive diagnostic likelihood ratio, LR− negative diagnostic likelihood ratio