Table 3 Performance metrics of the best tabular, image model and ensemble models using majority voting and average strategies. The results are the averages and standard deviations of the threefold validation.

Model	AUROC	F1-score	PPV	NPV	Sensitivity	Specificity
Tabular data model (AdaBoost)	0.738 ± 0.051	0.736 ± 0.024	0.734 ± 0.024	0.790 ± 0.019	0.746 ± 0.022	0.867 ± 0.019
Image model (Inception femur)	0.485 ± 0.019	0.605 ± 0.015	0.596 ± 0.017	0.703 ± 0.006	0.668 ± 0.003	0.908 ± 0.022
Image-based model (Inception femur, abdomen and head) Max voting	0.470 ± 0.016	0.576 ± 0.016	0.572 ± 0.010	0.693 ± 0.007	0.582 ± 0.023	0.718 ± 0.040
Image-based model (Inception femur, abdomen and head) Mean voting	0.477 ± 0.001	0.578 ± 0.006	0.587 ± 0.087	0.698 ± 0.001	0.694 ± 0.003	0.991 ± 0.003
AdaBoost + Inception femur, abdomen and head Max voting	0.621 ± 0.053	0.638 ± 0.043	0.679 ± 0.036	0.792 ± 0.033	0.624 ± 0.045	0.623 ± 0.047
AdaBoost + Inception femur, abdomen and head Mean voting	0.500 ± 0.001	0.578 ± 0.006	0.587 ± 0.087	0.698 ± 0.001	0.694 ± 0.003	0.991 ± 0.003
Final classification model (AdaBoost + Inception femur) Max voting	0.705 ± 0.049	0.716 ± 0.028	0.719 ± 0.027	0.802 ± 0.018	0.714 ± 0.029	0.783 ± 0.024
Final classification model (AdaBoost + Inception femur) Mean voting	0.522 ± 0.016	0.608 ± 0.012	0.606 ± 0.011	0.705 ± 0.005	0.677 ± 0.004	0.924 ± 0.022

Results are presented as mean ± standard deviation. These values were obtained by threefold cross-validation. Columns include: AUROC: area under the receiver operating curve; PPV: positive predictive value; NPV: negative predictive value.

Quick links

Search