Skip to main content

Table 2 Performances of the artificial intelligence system and ultrasound radiologists in diagnosing thyroid cancer

From: Automatic thyroid nodule recognition and diagnosis in ultrasound imaging with the YOLOv2 neural network

  Size, cm AI system Radiologist p value
Sensitivity, % < 1.0 88.46 93.08 0.284
1.0–4.0 94.34 95.28 1.000
≥ 4.0 66.67 83.33 1.000
All 90.5 93.80 0.237
Specificity, % < 1.0 57.14 14.29 0.009
1.0–4.0 95.74 89.36 0.435
≥ 4.0 100 97.56 1.000
All 89.91 77.98 0.026
PPV, % < 1.0 92.74 87.05 0.156
1.0–4.0 98.04 95.28 0.446
≥ 4.0 100 83.33 1.000
All 95.22 90.44 0.053
NPV, % < 1.0 44.44 25 0.305
1.0–4.0 88.24 89.36 1.000
≥ 4.0 95.35 97.56 1.000
All 80.99 85 0.477
Accuracy, % < 1.0 84.11 82.12 0.759
1.0–4.0 94.77 93.46 0.809
≥ 4.0 95.74 95.74 1.000
All 90.31 88.89 0.621
  1. PPV positive predictive value, NPV negative predictive value