Skip to main content

Table 2 Performance of the DCNN models on the internal validation and external testing datasets

From: Automated detection of myopic maculopathy from color fundus photographs using deep convolutional neural networks

Model

Accuracy (%)

κ

Sensitivity (%, 95% CI)

Specificity (%, 95% CI)

AUC (95% CI)

PM

TF

PM

TF

PM

TF

Internal validation

 Qingdao Eye Hospital North Branch of Shandong First Medical University (Camera: Canon, Syseye)

  DCNN-DS

96.5

0.922

96.6 (94.1–98.1)

94.7 (91.8–96.6)

99.3 (98.8–99.6)

97.0 (96.2–97.6)

0.998 (0.996–0.999)

0.989 (0.984–0.992)

  DCNN-P

93.1

0.862

93.2 (90.1–95.5)

93.1 (90.0–95.3)

98.8 (98.2–99.1)

94.1 (93.1–95.0)

0.995 (0.992–0.997)

0.978 (0.972–0.983)

  DCNN-O

90.5

0.800

88.8 (85.1–91.7)

91.9 (88.6–94.3)

97.9 (97.2–98.4)

91.8 (90.6–92.9)

0.991 (0.987–0.994)

0.968 (0.961–0.974)

 Rongcheng Eye Hospital (Camera: Zeiss)

  DCNN-DS

96.8

0.928

97.8 (95.6–99.0)

91.4 (87.6–94.1)

99.0 (98.4–99.3)

97.7 (97.0–98.3)

0.998 (0.995–0.999)

0.983 (0.977–0.987)

  DCNN-P

93.2

0.853

93.3 (90.1–95.5)

90.7 (86.9–93.6)

98.7 (98.0–99.1)

94.1 (93.0–95.0)

0.997 (0.994–0.999)

0.972 (0.965–0.978)

  DCNN-O

90.3

0.796

90.8 (87.3–93.5)

89.2 (85.2–92.3)

97.6 (96.8–98.2)

92.2 (91.0–93.3)

0.992 (0.988–0.995)

0.962 (0.954–0.969)

 Qilu Hospital of Shandong University (Qingdao) (Camera: Canon)

  DCNN-DS

95.6

0.903

93.8 (88.2–96.9)

92.6 (86.8–96.0)

99.3 (98.5–99.7)

96.4 (94.9–97.5)

0.997 (0.991–0.999)

0.984 (0.974–0.990)

  DCNN-P

93.6

0.862

92.4 (86.5–96.0)

92.6 (86.8–96.0)

99.0 (98.0–99.5)

94.2 (92.4–95.6)

0.995 (0.989–0.998)

0.982 (0.972–0.988)

  DCNN-O

90.4

0.798

83.4 (76.2–88.9)

93.2 (87.6–96.5)

98.3 (97.2–99.0)

91.4 (89.3–93.1)

0.989 (0.981–0.994)

0.966 (0.953–0.975)

 No. 971 Hospital of the People’s Liberation Army (Camera: Topcon)

  DCNN-DS

96.9

0.928

94.3 (88.1–97.5)

97.5 (92.4–99.4)

99.3 (98.3–99.7)

97.3 (95.9–98.3)

0.989 (0.980–0.994)

0.981 (0.970–0.988)

  DCNN-P

91.6

0.811

90.2 (83.1–94.6)

89.3 (82.1–94.0)

98.5 (97.4–99.2)

92.6 (90.5–94.3)

0.991 (0.983–0.995)

0.962 (0.948–0.972)

  DCNN-O

91.0

0.802

88.5 (81.2–93.4)

91.8 (85.1–95.8)

97.9 (96.7–98.8)

92.0 (89.9–93.7)

0.988 (0.979–0.993)

0.968 (0.955–0.977)

 Overall

  DCNN-DS

96.5

0.922

96.4 (95.0–97.4)

93.6 (91.9–95.0)

99.2 (98.9–99.4)

97.2 (96.8–97.6)

0.997 (0.995–0.998)

0.985 (0.982–0.988)

  DCNN-P

93.0

0.849

92.8 (91.0–94.2)

91.8 (89.9–93.4)

98.7 (98.4–99.0)

93.9 (93.3–94.5)

0.996 (0.994–0.997)

0.975 (0.971–0.978)

  DCNN-O

90.5

0.799

88.8 (86.6–90.6)

91.2 (89.2–92.8)

97.8 (97.5–98.2)

91.9 (91.2–92.6)

0.991 (0.989–0.993)

0.966 (0.962–0.970)

External testing

 Shandong Eye Hospital of Shandong First Medical University (Camera: Topcon)

  DCNN-DS

96.3

0.922

93.3 (90.6–95.2)

98.8 (98.1–99.2)

99.6 (99.5–99.8)

95.6 (95.0–96.1)

0.998 (0.997–0.999)

0.986 (0.983–0.988)

  DCNN-P

93.9

0.872

88.1 (84.9–90.8)

92.5 (91.1–93.7)

98.7 (98.4–98.9)

94.5 (93.8–95.1)

0.995 (0.993–0.996)

0.972 (0.968–0.976)

  DCNN-O

92.5

0.844

88.7 (85.6–91.3)

94.7 (93.5–95.7)

98.9 (98.7–99.2)

91.9 (91.1–92.6)

0.996 (0.994–0.997)

0.970 (0.966–0.974)

 Qingdao Eye Hospital of Shandong First Medical University (Camera: Kowa)

  DCNN-DS

93.0

0.797

91.0 (88.5–93.0)

92.8 (90.6–94.5)

98.7 (98.4–98.9)

94.1 (93.5–94.6)

0.994 (0.992–0.995)

0.970 (0.966–0.974)

  DCNN-P

92.3

0.772

79.8 (76.5–82.8)

87.1 (84.5–89.4)

98.5 (98.1–98.7)

93.8 (93.2–94.4)

0.990 (0.988–0.992)

0.967 (0.963–0.971)

  DCNN-O

90.6

0.731

74.5 (70.9–77.7)

89.0 (86.5–91.1)

98.2 (97.9–98.5)

91.7 (91.1–92.4)

0.987 (0.984–0.989)

0.960 (0.955–0.964)

  1. AUC = area under the curve; CI = confidence interval; PM = pathologic myopia; TF = tessellated fundus