Table 5 Performance metrics for the deep neural networks in tuberculosis detection (o.o.d. evaluation)^a

From: Improving deep neural network generalization and robustness to background bias via layer-wise relevance propagation optimization

Model and metric	Normal	Tuberculosis	Mean (macro-average)
ISNet precision	0.744 ± 0.045	0.734 ± 0.043	0.739 ± 0.044
U-Net+DenseNet121 precision	0.63 ± 0.061	0.573 ± 0.043	0.601 ± 0.052
DenseNet121 precision	0.578 ± 0.055	0.564 ± 0.046	0.571 ± 0.05
Multi-task U-Net precision	0.515 ± 0.048	0.539 ± 0.053	0.527 ± 0.05
AG-Sononet precision	0.731 ± 0.06	0.599 ± 0.041	0.665 ± 0.05
Extended GAIN precision	0.576 ± 0.04	0.766 ± 0.06	0.671 ± 0.05
RRR precision	0.663 ± 0.049	0.664 ± 0.046	0.663 ± 0.048
Vision Transformer (ViT-B/16) precision	0.52 ± 0.044	0.56 ± 0.059	0.54 ± 0.052
ISNet recall	0.714 ± 0.046	0.762 ± 0.042	0.738 ± 0.044
U-Net+DenseNet121 recall	0.409 ± 0.05	0.768 ± 0.042	0.589 ± 0.046
DenseNet121 recall	0.479 ± 0.05	0.659 ± 0.047	0.569 ± 0.048
Multi-task U-Net recall	0.586 ± 0.05	0.468 ± 0.05	0.527 ± 0.05
AG-Sononet recall	0.406 ± 0.05	0.855 ± 0.035	0.631 ± 0.042
Extended GAIN recall	0.883 ± 0.032	0.372 ± 0.048	0.627 ± 0.04
RRR recall	0.642 ± 0.049	0.685 ± 0.046	0.663 ± 0.048
Vision Transformer (ViT-B/16) recall	0.679 ± 0.047	0.395 ± 0.049	0.537 ± 0.048
ISNet F1-Score	0.729 ± 0.046	0.748 ± 0.043	0.738 ± 0.044
U-Net+DenseNet121 F1-Score	0.496 ± 0.056	0.656 ± 0.044	0.576 ± 0.05
DenseNet121 F1-Score	0.524 ± 0.052	0.608 ± 0.047	0.566 ± 0.05
Multi-task U-Net F1-Score	0.548 ± 0.049	0.501 ± 0.052	0.524 ± 0.05
AG-Sononet F1-Score	0.522 ± 0.057	0.704 ± 0.04	0.613 ± 0.048
Extended GAIN F1-Score	0.697 ± 0.039	0.501 ± 0.056	0.599 ± 0.048
RRR F1-Score	0.652 ± 0.049	0.674 ± 0.046	0.663 ± 0.048
Vision Transformer (ViT-B/16) F1-Score	0.589 ± 0.046	0.463 ± 0.054	0.526 ± 0.05
ISNet AUC	0.809 ± 0.031	0.809 ± 0.031	0.809 ± 0.031
U-Net+DenseNet121 AUC	0.667 ± 0.039	0.667 ± 0.039	0.667 ± 0.039
DenseNet121 AUC	0.576 ± 0.04	0.576 ± 0.04	0.576 ± 0.04
Multi-task U-Net AUC	0.549 ± 0.041	0.549 ± 0.041	0.549 ± 0.041
AG-Sononet AUC	0.717 ± 0.037	0.717 ± 0.037	0.717 ± 0.037
Extended GAIN AUC	0.676 ± 0.038	0.676 ± 0.038	0.676 ± 0.038
RRR AUC	0.728 ± 0.036	0.728 ± 0.036	0.728 ± 0.036
Vision Transformer (ViT-B/16) AUC	0.558 ± 0.041	0.558 ± 0.041	0.558 ± 0.041

^aThe cells display the metrics’ mean and 95% confidence intervals.

Back to article page

Table 5 Performance metrics for the deep neural networks in tuberculosis detection (o.o.d. evaluation)^a

Search

Quick links