Table 3 Performance metrics with Test C 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

0.795 (0.782–0.808)

0.667 (0.631–0.703)

0.930 (0.902–0.958)

0.404 (0.308–0.500)

0.612 (0.578–0.646)

0.856 (0.829–0.883)

SP-CNN(Neu)

0.769 (0.748–0.790)

0.642 (0.614–0.670)

0.922 (0.897–0.947)

0.362 (0.284–0.440)

0.593 (0.569–0.616)

0.826 (0.800–0.852)

SP-CNN(Flx)

0.773 (0.747–0.798)

0.651 (0.635–0.667)

0.900 (0.867–0.933)

0.402 (0.340–0.464)

0.602 (0.585–0.619)

0.806 (0.778–0.883)

SP-CNN(Ext)

0.742* (0.726–0.758)

0.661 (0.638–0.684)

0.886 (0.848–0.924)

0.436 (0.366–0.506)

0.612 (0.591– 0.634)

0.798 (0.760–0.836)

VGG19

MP-CNN

0.772 (0.743–0.801)

0.683 (0.657–0.709)

0.920 (0.879–0.961)

0.446 (0.377–0.515)

0.625 (0.602–0.649)

0.857 (0.800–0.914)

SP-CNN(Neu)

0.776* (0.754–0.798)

0.667 (0.642–0.692)

0.926 (0.883–0.969)

0.408 (0.329–0.486)

0.612 (0.589–0.648)

0.883 (0.838–0.928)

SP-CNN(Flx)

0.743 (0.707–0.780)

0.648 (0.624–0.672)

0.920 (0.872–0.968)

0.376 (0.315–0.437)

0.596 (0.579–0.635)

0.859 (0.808–0.911)

SP-CNN(Ext)

0.721* (0.695–0.747)

0.665 (0.635–0.695)

0.858 (0.794–0.922)

0.472 (0.395–0.549)

0.620 (0.595–0.645)

0.780 (0.714–0.847)

VGG16

MP-CNN

0.775 (0.743–0.807)

0.667 (0.636–0.698)

0.952 (0.935–0.969)

0.382 (0.313–0.451)

0.608 (0.582–0.633)

0.890 (0.856–0.924)

SP-CNN(Neu)

0.769 (0.727–0.811)

0.630 (0.600–0.660)

0.984 (0.971–0.997)

0.276 (0.207–0.345)

0.577 (0.556–0.599)

0.951 (0.917–0.984)

SP-CNN(Flx)

0.743 (0.695–0.790)

0.641 (0.606–0.675)

0.903 (0.883–0.923)

0.384 (0.322–0.446)

0.590 (0.566–0.615)

0.796 (0.735–0.857)

SP-CNN(Ext)

0.725 (0.689–0.761)

0.670 (0.652–0.688)

0.902 (0.876–0.928)

0.438 (0.394–0.482)

0.617 (0.601–0.632)

0.820 (0.784–0.856)

EfficientNet-B1

MP-CNN

0.748 (0.727–0.768)

0.634 (0.624–0.644)

0.944 (0.934–0.954)

0.324 (0.306–0.342)

0.583 (0.576– 0.590)

0.853 (0.830–0.876)

SP-CNN(Neu)

0.726 (0.698–0.754)

0.604 (0.581–0.627)

0.920 (0.896–0.944)

0.288 (0.242–0.334)

0.564 (0.548–0.580)

0.783 (0.733–0.834)

SP-CNN(Flx)

0.726* (0.701–0.751)

0.631 (0.603–0.659)

0.946 (0.938–0.954)

0.316 (0.253–0.379)

0.581 (0.561–0.601)

0.852 (0.840–0.864)

SP-CNN(Ext)

0.700 (0.669–0.732)

0.623 (0.597–0.649)

0.888 (0.858–0.918)

0.358 (0.293–0.423)

0.581 (0.562–0.601)

0.762 (0.728–0.796)

  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