Table 2 Performance of the deep learning models vs. general ophthalmologists in the test sets.

From: Deep learning for detecting retinal detachment and discerning macular status using ultra-widefield fundus images

 

Sensitivity (95% CI)

Specificity (95% CI)

Accuracy (95% CI)

Retinal detachment

  Ophthalmologist A

95.4% (93.0–97.8)

99.8% (99.6–100)

99.0% (98.5–99.5)

  Ophthalmologist B

94.4% (91.7–97.1)

99.1% (98.6–99.6)

98.2% (97.5–98.9)

  The first model

96.1% (93.9–98.3)

99.6% (99.3–99.9)

98.9% (98.4–99.4)

Macula-on detachment

  Ophthalmologist A

91.3% (84.8–97.8)

92.4% (88.4–96.4)

92.1% (88.7–95.5)

  Ophthalmologist B

86.3% (78.2–94.4)

87.1% (81.9–92.3)

86.8% (82.4–91.2)

  The second model

93.8% (88.3–99.3)

90.9% (86.6–95.2)

91.7% (88.2–95.2)

  1. A general ophthalmologist with 5 years of working experience at a physical examination centre, B general ophthalmologist with 3 years of working experience at a physical examination centre, CI confidence interval
Back to article page