A machine learning model that classifies breast cancer pathologic complete response on MRI post-neoadjuvant chemotherapy

Table 2 Performance of the RFE-RF classifier trained using model 1 and model 2 for predicting a pCR. P values are derived from comparison of the ROC curves computed for the cross-validation and test sets for model 1 and model 2

	Model 1		Model 2
	Radiomics		Radiomics and molecular subtype
	*Training*	*Testing*	*Training*	*Testing*
AUROC 95% CI	0.72 (0.64, 0.79)	0.83 (0.71, 0.94)	0.80 (0.72, 0.87)	0.78 (0.62, 0.94)
Sensitivity or TPR (no-pCR)	0.73 (0.65, 0.79)	0.77 (0.61, 0.88)	0.78 (0.70, 0.84)	0.79 (0.64, 0.90)
Specificity or TNR (pCR)	0.64 (0.51, 0.76)	0.69 (0.39, 0.91)	0.69 (0.56, 0.80)	0.69 (0.39, 0.91)
PPV	0.84 (0.77, 0.90)	0.89 (0.75, 0.97)	0.87 (0.80, 0.92)	0.89 (0.75, 0.97)
NPV	0.47 (0.36, 0.58)	0.47 (0.24, 0.71)	0.54 (0.42, 0.65)	0.50 (0.26, 0.74)
	P value = 0.1		P value = 0.9

Abbreviations: AUC area under the receiver operating characteristic curve, pCR pathologic complete response, TPR true positive rate, TNR true negative rate
*P value < 0.05

ISSN: 1465-542X