Table 2.

Comparison of LNM predictions in different cohorts using the MMD-DL model

Cohorts	AUC	Accuracy	Sensitivity	Specificity	PPV	NPV
Pre-training (n = 488)	0.91 [0.89, 0.94]	85 (351/412) [81, 88]	83 (180/216) [78, 88]	87 (171/196) [82, 92]	88 (180/205) [83, 91]	83 (171/207) [78, 87]
Training (n = 109)	0.85 [0.78, 0.92]	74 (81/109) [65, 82]	72 (36/50) [57, 83]	76 (45/59) [63, 86]	72 (36/50) [57, 83]	76 (45/59) [63, 86]
Test (n = 39)	0.85 [0.73, 0.97]	80 (31/39) [64, 91]	63 (12/19) [39, 83]	95 (19/20) [73, 99]	92 (12/13) [62, 99]	73 (19/26) [52, 88]
Validation (n = 70)	0.81 [0.73, 0.89]	80 (56/70) [69, 89]	69 (22/32) [50, 84]	89 (34/38) [75, 97]	85 (22/26) [68, 93]	77 (34/44) [69, 89]
Hospital 1 (n = 23)	0.78 [0.57, 0.99]	83 (19/23) [61, 95]	89 (8/9) [51, 99]	79 (11/14) [49, 94]	73 (8/11) [39, 93]	92 (11/12) [60, 99]
Hospital 2 (n = 28)	0.77 [0.59, 0.95]	79 (22/28) [54, 94]	89 (8/9) [51, 99]	74 (14/19) [51, 99]	62 (8/13) [32, 85]	93 (14/15) [66, 100]
Hospital 3 (n = 19)	0.84 [0.67, 1.00]	79 (15/19) [59, 92]	60 (6/10) [27, 86]	100 (9/9) [63, 100]	100 (6/6) [52, 100]	70 (9/13) [39, 90]

Unless otherwise specified, data are percentages, with the number of patients in parentheses and 95% confidence intervals in brackets. AUC, area under the receiver operating characteristic curve; PPV, positive predictive value; NPV, negative predictive value