Automatic segmentation and classification of breast lesions through identification of informative multiparametric PET/MRI features

European Radiology Experimental

Table 5 Classification results for differentiation of malignant and benign lesions for manually annotated lesions and automatic segmented lesions using automatic feature selection

Feature selection method	Manual annotation			Automatic segmentation
	AUC (mean ± SD)	Sensitivity/specificity	Number of features	AUC (mean ± SD)	Sensitivity/specificity	Number of features
GI LOOCV	0.949 ± 0.019	0.920/0.868	4	0.771 ± 0.040	0.961/0.482	100
GI LOOCV without PET	0.946 ± 0.002	0.924/0.859	4	0.771 ± 0.040	0.972/0.486	75
GI LOOCV without DWI	0.949 ± 0.015	0.915/0.873	4	0.754 ± 0.035	0.983/0.427	75
GI LOOCV without DWI, without PET	0.944 ± 0.018	0.922/0.868	4	0.755 ± 0.033	0.976/0.409	75
mRMR LOOCV	0.978 ± 0.008	0.946/0.936	2	0.858 ± 0.013	0.941/0.773	3
mRMR LOOCV w/o PET	0.975 ± 0.010	0.957/0.918	2	0.856 ± 0.018	0.948/0.736	3
mRMR LOOCV w/o DWI	0.977 ± 0.006	0.954/0.950	2	0.857 ± 0.017	0.943/0.745	3
mRMR LOOCV w/o DWI, PET	0.973 ± 0.010	0.950/0.927	2	0.861 ± 0.009	0.941/0.755	3

AUC area under the curve at receiver operating characteristic analysis, DCE-MRI dynamic contrast-enhanced magnetic resonance imaging, DWI diffusion-weighted imaging, GI Gini importance, LOOCV leave-one-out cross-validation, mRMR minimum-redundancy-maximum-relevance, PET positron emission tomography, SD standard deviation. Values presented in bold are the highest values