Table 3.

Diagnostic accuracy for detecting pelvic lymph node metastasis based on selected magnetic resonance (MR) imaging features.

Parameters	n	TP	TN	FP	FN	Sensitivity	Specificity	Accuracy
Region basis (Training)
RadSignature	330	22	267	40	1	95.7% (78.1–99.9%)	87.0% (82.7–90.5%)	87.6% (83.5–90.9%)
RadScore	330	22	248	59	1	95.7% (78.1–99.9%)	80.8% (75.9–85.0%)	81.8% (77.2–85.8%)
ADC	330	21	91	216	2	91.3% (72.0–98.9%)	29.6% (24.6–35.1%) *	33.9% (28.8–39.3%) *
SA	330	16	213	94	7	69.6% (47.1–86.8%) *	69.4% (63.9–74.5%) *	69.4% (64.1–74.3%) *
Region basis (Testing)
RadSignature	142	8	114	18	2	80.0% (44.4–97.5%)	86.4% (79.3–91.7%)	85.9% (79.1–91.2%)
RadScore	142	8	106	26	2	80.0% (44.4–97.5%)	80.3% (72.5–86.7%)	80.3% (72.8–86.5%)
ADC	142	7	32	100	3	70.0% (34.8–93.3%)	24.2% (17.2–32.5%) *	27.5% (20.3–35.6%) *
SA	142	7	85	47	3	70.0% (34.8–93.3%)	64.4% (55.6–72.5%) *	64.8% (56.3–72.6%) *
Patient basis (Training)
RadSignature	165	15	119	30	1	93.8% (69.8–99.8%)	79.9% (72.5–86.0%)	81.2% (74.4–86.9%)
RadScore	165	15	105	44	1	93.8% (69.8–99.8%)	70.5% (62.5–77.7%)	72.7% (65.3–79.4%)
ADC	165	15	25	124	1	93.8% (69.8–99.8%)	16.8% (11.2–23.8%) *	24.2% (17.9–31.5%) *
SA	165	12	80	69	4	75.0% (47.6–92.7%)	53.7% (45.3–61.9%) *	55.8% (47.8–63.5%) *
Patient basis (Testing)
RadSignature	71	6	50	14	1	85.7% (42.1–99.6%)	78.1% (66.0–87.5%)	78.9% (67.6–87.7%)
RadScore	71	6	44	20	1	85.7% (42.1–99.6%)	68.8% (55.9–79.8%)	70.4% (58.4–80.7%)
ADC	71	6	9	55	1	85.7% (42.1–99.6%)	14.1% (6.6–25.0%) *	21.1% (12.3–32.4%) *
SA	71	5	28	36	2	71.4% (29.0–96.3%)	43.8% (31.4–56.7%) *	46.5% (34.5–58.7%) *

Data in parentheses are 95% confidence intervals. TP = true positive, TN = true negative, FP = false positive, FN = false negative, PPV = positive predictive value, NPV = negative predictive value, SA = short axis, ADC = mean apparent diffusion coefficient value of the lymph node. * p < 0.05 McNemar test, as compared with the RadSignature model.