Skip to main content
. Author manuscript; available in PMC: 2020 Dec 1.
Published in final edited form as: J Magn Reson Imaging. 2019 Nov 25;52(6):1607–1619. doi: 10.1002/jmri.27001

TABLE 2.

Diagnostic Performance of Deep Learning Methods for Detecting Fractures

Machine performance
Fracture site Dataset size CNN used AUC Sensitivity Specificity Accuracy
Hip 33 805 GooLeNet 0.98 NS NS 94%
Hip 34 3,605 DenseNet 0.98 98% 84% 91%
Hip 35 3,346 VGG-16 NS 94% 97% 96%
Shoulder 37 1,891 ResNet 0.99 99% 97% 95%
Wrist 40 7,356 ResNet 0.90 98% 73% NS
Wrist 38 1,389 Inception 0.95 90% 88% NS
Wrist 39 256,000 VGG-16 NS NS NS 82%
Ankle 41 596 Xception NS 73% 76% 75%
All Sites 36 135,409 U-Net 0.99 94% 95% NS