Intra- and peri-tumoral radiomics based on dynamic contrast-enhanced MRI for prediction of benign disease in BI-RADS 4 breast lesions: a multicentre study

Table 3 Performance of the best models established by using the features extracted from seven different ROIs

Models	Algorithm	Cohort	AUC	95% CI	Sensitivity	Specificity	Accuracy	PPV	NPV
Intra	SVM	Training	0.898	0.867–0.928	0.793	0.821	0.732	0.798	0.817
		Internal test	0.724	0.633–0.816	0.723	0.712	0.719	0.819	0.587
		External test	0.797	0.667–0.927	0.767	0.682	0.731	0.767	0.682
Peri1	SVM	Training	0.801	0.755–0.846	0.948	0.291	0.600	0.543	0.864
		Internal test	0.793	0.719–0.867	0.968	0.250	0.712	0.700	0.812
		External test	0.815	0.696–0.934	0.933	0.227	0.635	0.622	0.714
Peri2	SVM	Training	0.910	0.881–0.939	0.885	0.760	0.819	0.766	0.882
		Internal test	0.783	0.708–0.858	0.851	0.519	0.733	0.762	0.659
		External test	0.823	0.704–0.942	0.867	0.773	0.827	0.839	0.810
Peri3	LightGBM	Training	0.817	0.775–0.860	0.649	0.791	0.724	0.734	0.718
		Internal test	0.748	0.663–0.833	0.628	0.769	0.678	0.831	0.533
		External test	0.788	0.659–0.917	0.700	0.773	0.731	0.808	0.654
Comb1	SVM	Training	0.907	0.879–0.936	0.764	0.862	0.816	0.831	0.805
		Internal test	0.737	0.651–0.822	0.670	0.712	0.685	0.808	0.544
		External test	0.835	0.715–0.955	0.967	0.545	0.788	0.744	0.923
Comb2	SVM	Training	0.916	0.886–0.946	0.862	0.801	0.830	0.794	0.867
		Internal test	0.828	0.763–0.893	0.830	0.615	0.753	0.796	0.667
		External test	0.844	0.713–0.975	0.967	0.682	0.846	0.806	0.937
Comb3	SVM	Training	0.856	0.817–0.894	0.960	0.332	0.627	0.560	0.903
		Internal test	0.788	0.712–0.865	0.968	0.173	0.685	0.679	0.750
		External test	0.812	0.696–0.928	0.967	0.227	0.654	0.630	0.833

* Intra represents the model based on features extracted from the intra-tumoral region. Peri1, Peri2 and Peri3 represent the models based on features extracted from the peri-tumoral regions of 1 mm, 2 mm and 3 mm, respectively. Comb1, Comb2 and Comb3 represent the models based on features extracted from the intra-tumoral region combined with the peri-tumoral regions of 1 mm, 2 mm and 3 mm, respectively. Abbreviations: AUC, the area under curve; CI, confidence interval; PPV, Positive predictive value; NPV, Negative predictive value; SVM, Support vector machine; GBM, Gradient Boosting Machine

ISSN: 1748-717X