Fig. 4: Feature importance scores.
From: Deep learning predicts all-cause mortality from longitudinal total-body DXA imaging
Scores are change in area under the receiver operating characteristic (AUROC) compared to the full model and error bars show 95% confidence intervals calculated using bootstrapping. a Ablation study results on test set (n = 3720) for single-record model. b Ablation study results on test set (n = 639) for sequence model.
