Table 3 Accuracy of Google teachable machine algorithms (external validation; cutoff level = 0.5).

	AUC	Sen	Sp	PPV	NPV	LR+	LR−	OA	F1 score
Validation dataset 1 (TB vs. normal)
Model 1	0.800	0.65	0.89	0.85	0.72	5.90	0.39	0.77	0.73
Model 2	0.902	0.83	0.83	0.82	0.83	4.88	0.20	0.83	0.82
Model 3	0.951	0.88	0.95	0.94	0.88	17.6	0.12	0.91	0.91
Validation dataset 1 (TB vs. normal and other abnormality)
Model 1	0.720	0.65	0.73	0.54	0.80	2.40	0.47	0.70	0.59
Model 2	0.656	0.83	0.44	0.42	0.83	1.48	0.38	0.56	0.56
Model 3	0.758	0.88	0.52	0.47	0.89	1.83	0.23	0.64	0.61
Validation dataset 2 (TB vs. normal)
Model 1	0.795	0.68	0.93	0.93	0.65	9.71	0.34	0.78	0.78
Model 2	0.917	0.86	0.90	0.92	0.81	8.60	0.15	0.87	0.89
Model 3	0.975	0.86	1.0	1.0	0.82	Infinity	0.14	0.91	0.92
Validation dataset 2 (TB vs. normal and other abnormality)
Model 1	0.752	0.68	0.81	0.76	0.73	3.57	0.39	0.74	0.72
Model 2	0.718	0.86	0.58	0.65	0.82	2.04	0.24	0.71	0.74
Model 3	0.828	0.86	0.65	0.69	0.83	2.45	0.21	0.75	0.76

AUC area under curve, LR likelihood ratio, NPV negative predictive value, OA overall accuracy, PPV positive predictive value, Sen sensitivity, Sp specificity.

Quick links

Search