Table 1 The diagnostic performance for comparisons. All models are being trained using ImageNet_1k pretrained weights. The performance improvement is significant for \(p<0.05\), highly significant for \(p<0.01\), then extremely significant for \(p<0.001\).
From: Multimodal-based auxiliary diagnosis for pediatric community acquired pneumonia
CXR Model | Accuracy (95% CI) | Precision (95% CI) | Recall (95% CI) | F1-Score (95% CI) | GFLOPs | t-test |
|---|---|---|---|---|---|---|
MNv3-L | 0.829 (0.821 - 0.837) | 0.838 (0.813 - 0.863) | 0.819 (0.787 - 0.852) | 0.827 (0.818 - 0.837) | 2.973 | \(p<0.01\) |
Dense-121 | 0.834 (0.822 - 0.847) | 0.845 (0.824 - 0.865) | 0.827 (0.782 - 0.872) | 0.834 (0.817 - 0.851) | 15.130 | \(p<0.05\) |
ViT_b_16 | 0.796 (0.782 - 0.809) | 0.767 (0.728 - 0.806) | 0.847 (0.806 - 0.889) | 0.808 (0.800 - 0.816) | 58.721 | \(p<0.001\) |
MaxViT_t | 0.823 (0.811 - 0.834) | 0.861 (0.830 - 0.892) | 0.774 (0.738 - 0.810) | 0.814 (0.800 - 0.828) | 30.117 | \(p<0.01\) |
STv2_s | 0.818 (0.807 - 0.830) | 0.823 (0.803 - 0.843) | 0.813 (0.785 - 0.841) | 0.817 (0.805 - 0.830) | 15.568 | \(p<0.01\) |
Effv2_m | 0.849 (0.839 - 0.859) | 0.869 (0.844 - 0.894) | 0.826 (0.800 - 0.851) | 0.846 (0.836 - 0.856) | 28.410 | \(p>0.5\) |
ResNet-50 | 0.846 (0.838 - 0.854) | 0.837 (0.809 - 0.866) | 0.863 (0.830 - 0.897) | 0.849 (0.842 - 0.856) | 21.585 | - |
