Contrast-enhanced CT radiomics for preoperative prediction of stage in epithelial ovarian cancer: a multicenter study

Table 3 The results of five machine learning algorithms utilized for constructing radiomics models

Model	AUC (95% CI)	Sensitivity	Specificity	Accuracy	F1 score
LightGBM
Training	0.83 (0.75–0.91)	0.75	0.77	0.75	0.79
Internal validation	0.80 (0.67–0.93)	0.76	0.65	0.72	0.77
External validation	0.68 (0.52–0.84)	0.68	0.76	0.71	0.73
LR
Training	0.80 (0.71–0.89)	0.75	0.77	0.76	0.82
Internal validation	0.80 (0.66–0.93)	0.72	0.77	0.74	0.78
External validation	0.51 (0.34–0.67)	0.46	0.48	0.47	0.40
SVM
Training	0.75 (0.65–0.86)	0.88	0.59	0.77	0.81
Internal validation	0.83 (0.71–0.96)	0.72	0.88	0.78	0.81
External validation	0.54 (0.37–0.70)	0.64	0.38	0.53	0.44
DT
Training	0.95 (0.91–0.98)	0.75	1.00	0.84	0.89
Internal validation	0.62 (0.45–0.79)	0.97	0.41	0.76	0.84
External validation	0.60 (0.44–0.75)	0.86	0.33	0.63	0.73
RF
Training	0.99 (0.98-1.00)	0.93	1.00	0.95	0.95
Internal validation	0.75 (0.60–0.90)	0.76	0.71	0.74	0.78
External validation	0.67 (0.51–0.83)	0.54	0.81	0.65	0.69

LightGBM, light gradient boosting machine; LR, logistic regression; SVM, support vector machine; DT, decision tree; RF, random forest; AUC, area under the curve; 95% CI, 95% confidence interval

ISSN: 1471-2407