Table 5 Performance of Deep learning and Radiologist in Diagnosing PLC injury.

From: Deep learning system for automated detection of posterior ligamentous complex injury in patients with thoracolumbar fracture on MRI

 

Sensitivity

Specificity

PPV

NPV

AUROC

DL algorithm

0.820 (0.686–0.914)

0.940 (0.835–0.988)

0.932 (0.819–0.976)

0.839 (0.742–0.905)

0.916 (0.844–0.963)

MSK radiologist (R1)

0.920 (0.808–0.978)

0.940 (0.835–0.988)

0.939 (0.836–0.979)

0.922 (0.821–0.968)

0.930 (0.861–0.971)

Radiology trainee (R2)

without DL-assistance

0.680 (0.533–0.805)

0.980 (0.894–0.999)

0.971 (0.829–0.996)

0.754 (0.671–0.821)

0.830 (0.742–0.898)

R2 with DL-assistance

0.880 (0.757–0.955)

0.960 (0.863–0.995)

0.957 (0.849–0.989)

0.889 (0.790–0.944)

0.920 (0.848–0.965)

  1. †Data in the parentheses are 95% confidence intervals.
  2. ‡PLC, posterior ligamentous complex; DL, deep learning; PPV, positive predictive value; NPV, negative predictive value; AUROC, area under the curve of the receiver operating characteristics; MSK, musculoskeletal.
Back to article page