Table 2 Comparison of NCLS and radiologists
From: Deep learning approach for screening neonatal cerebral lesions on ultrasound in China
Sets | Group | SEN | SPE | PPV | NPV | F1 | AUC | P value |
|---|---|---|---|---|---|---|---|---|
Internal test set | NCLS | 87.50 (68.74–100.0) | 93.44 (89.73–96.70) | 53.85 (34.45–72.41) | 98.84 (97.02–100.0) | 66.67 (48.26–81.25) | 98.16 (95.89–99.65) | – |
Mid-level radiologists | 81.25 (74.69– 86.73) | 98.66 (98.05–99.11) | 84.12 (77.74–89.26) | 98.37 (97.71–98.87) | 82.66 (78.09–86.85) | 89.95 (86.90–92.75) | 0.0009 | |
Junior radiologists (w/o AI) | 87.50 (80.97– 92.42) | 85.12 (83.31–86.81) | 33.96 (29.15–39.03) | 98.73 (98.00–99.25) | 48.93 (43.76–54.29) | 86.31 (83.09–88.91) | 0.0019 | |
Junior radiologists (w. AI) | 97.22 (93.04– 99.24) | 89.37 (87.79–90.82) | 44.44 (38.87–50.12) | 99.73 (99.31–99.93) | 62.95 (55.90–66.53) | 93.30 (91.67–94.69) | 0.0097* | |
External test set | NCLS | 96.15 (86.94–100.0) | 92.73 (89.88–95.30) | 51.02 (36.36–65.91) | 99.67 (98.99–100.0) | 66.67 (52.17–77.78) | 94.44 (89.42–97.31) | – |
Mid-level radiologists | 71.68 (66.08– 76.83) | 97.55 (96.99–98.03) | 69.73 (64.13–74.93) | 97.76 (97.23–98.22) | 70.69 (66.67–74.60) | 84.61 (82.95–87.21) | 0.0004 | |
Junior radiologists (w/o AI) | 71.79 (65.56–77.46) | 92.36 (91.34–93.29) | 42.53 (37.60–47.57) | 97.56 (97.02–98.18) | 53.42 (48.47–57.62) | 82.08 (79.22–85.04) | 0.0019 | |
Junior radiologists (w. AI) | 91.03 (86.61– 94.36) | 91.25 (90.17–92.24) | 45.03 (40.49–49.64) | 99.23 (98.83–99.52) | 62.20 (55.67–64.35) | 91.14 (89.17–92.86) | 0.0039* |
