A hierarchical opportunistic screening model for osteoporosis using machine learning applied to clinical data and CT images

Table 2 Classification performance of each classifier on clinical data and CT images

Features		LR	SVM	ANN	RF	XGBoost	Stacking	N
DC	Training	0.798	0.800	0.795	0.836	0.828	0.819	2188
DC	Test	0.805	0.806	0.798	0.809	0.808	0.810
RLT	Training	0.696	0.725	0.702	0.856	0.828	0.819
RLT	Test	0.677	0.694	0.680	0.687	0.687	0.694
CD	Training	0.815	0.837	0.818	0.898	0.893	0.872
CD	Test	0.813	0.824	0.815	0.820	0.820	0.828
TFs	Training	0.970	0.971	0.942	0.989	0.976	0.976	268
TFs	Test	0.949	0.951	0.929	0.947	0.933	0.953
SFs	Training	0.869	0.882	0.829	0.945	0.905	0.892
SFs	Test	0.855	0.875	0.850	0.867	0.853	0.876
IFs*	Training	0.977	0.979	0.932	0.978	0.973	0.982
IFs*	Test	0.950	0.957	0.931	0.960	0.938	0.959

DC Demographic Characteristics; RLT Routine Laboratory Tests; CD Clinical Data; TFs Texture Features; SFs Shape Features; IFs Image Features
The performances of each classifier were evaluated by the mean of five repeated experiments
The highest values among the six classifiers for each feature set in test set were highlighted in bold
*Image features included texture and shape features

ISSN: 1471-2105