Table 1 Performance metrics with Test A for MP-CNN model vs SP-CNN model.

From: Multi-pose-based convolutional neural network model for diagnosis of patients with central lumbar spinal stenosis

Algorithm

Model

AUROC (95% CI)

Accuracy (95% CI)

Sensitivity (95% CI)

Specificity (95% CI)

PPV (95% CI)

NPV (95% CI)

ResNet50

MP-CNN

0.914 (0.909–0.918)

0.842 (0.837–0.846)

0.850 (0.817–0.883)

0.833 (0.794–0.872)

0.839 (0.812–0.866)

0.848 (0.826–0.860)

SP-CNN(Neu)

0.896* (0.890–0.902)

0.819 (0.812–0.827)

0.850 (0.828–0.872)

0.788 (0.758–0.818)

0.803 (0.784–0.822)

0.839 (0.824–0.854)

SP-CNN(Flx)

0.882* (0.875–0.888)

0.809 (0.796–0.823)

0.834 (0.811–0.857)

0.785 (0.773–0.797)

0.797 (0.786–0.807)

0.824 (0.804–0.845)

SP-CNN(Ext)

0.893* (0.890–0.897)

0.813 (0.800–0.825)

0.837 (0.808–0.867)

0.788 (0.756–0.820)

0.801 (0.779–0.822)

0.829 (0.807–0.850)

VGG19

MP-CNN

0.920 (0.916–0.923)

0.854 (0.845–0.863)

0.855 (0.836–0.873)

0.854 (0.837–0.872)

0.856 (0.842–0.871)

0.852 (0.844–0.860)

SP-CNN(Neu)

0.900* (0.894–0.906)

0.828 (0.821–0.836)

0.846 (0.815–0.878)

0.810 (0.787–0.833)

0.818 (0.805–0.831)

0.845 (0.828–0.862)

SP-CNN(Flx)

0.901* (0.898–0.904)

0.836 (0.833–0.839)

0.846 (0.840–0.851)

0.826 (0.818–0.834)

0.831 (0.825–0.837)

0.822 (0.809–0.835)

SP-CNN(Ext)

0.898* (0.891–0.905)

0.827 (0.820–0.834)

0.842 (0.827–0.857)

0.812 (0.797–0.827)

0.819 (0.809–0.830)

0.832 (0.814–0.850)

VGG16

MP-CNN

0.914 (0.910–0.917)

0.850 (0.843–0.857)

0.855 (0.846–0.863)

0.846 (0.834–0.857)

0.848 (0.839–0.858)

0.848 (0.826–0.860)

SP-CNN(Neu)

0.896* (0.891–0.901)

0.824 (0.816–0.832)

0.856 (0.828–0.877)

0.792 (0.779–0.805)

0.806 (0.798–0.814))

0.839 (0.824–0.854)

SP-CNN(Flx)

0.888* (0.882–0.894)

0.819 (0.814–0.824)

0.825 (0.808–0.843)

0.812 (0.805–0.819)

0.816 (0.813–0.819)

0.824 (0.804–0.845)

SP-CNN(Ext)

0.895* (0.889–0.901)

0.819 (0.813–0.825)

0.839 (0.816–0.863)

0.798 (0.786–0.811)

0.808 (0.803–0.814)

0.829 (0.807–0.850)

EfficientNet-B1

MP-CNN

0.905 (0.901–0.908)

0.839 (0.833–0.846)

0.840 (0.825–0.855)

0.839 (0.829–0.849)

0.841 (0.833 –0.848)

0.839 (0.827–0.851)

SP-CNN(Neu)

0.887* (0.875–0.898)

0.815 (0.804–0.825)

0.823 (0.802–0.843)

0.806 (0.789–0.824)

0.811 (0.799–0.824)

0.819 (0.803–0.835)

SP-CNN(Flx)

0.881* (0.870–0.891)

0.817 (0.803–0.832)

0.825 (0.811–0.839)

0.810 (0.793–0.827)

0.814 (0.799–0.830)

0.821 (0.807–0.835)

SP-CNN(Ext)

0.883* (0.877–0.889)

0.811 (0.804–0.818)

0.819 (0.795–0.842)

0.804 (0.790–0.819)

0.809 (0.801–0.817)

0.815 (0.798–0.833)

  1. The best values are in bold.
  2. Statistical analysis was performed using a paired t-test to compare the AUROC of each SP-CNN model with that of the MP-CNN model. An asterisk (*) means statistically significant difference (p < 0.05).
  3. AUROC area under the receiver operating characteristics curve, PPV positive predictive value, NPV negative predictive value, CI confidence interval, MP-CNN multi pose-based convolutional neural network model, SP-CNN single pose-based convolutional neural network, Neu neutralposture, Flx flexion posture, Ext extension posture.
Back to article page