Longitudinal Screening for Diabetic Retinopathy in a Nationwide Screening Program: Comparing Deep Learning and Human Graders
Table 2
The number of patients with sight-threatening diabetic retinopathy including the screening outcomes in the first and second screening determined by each modality.
Modality
Metric
First screening
Second screening
Difference (%)
value
DL
No. of patients
5,738
4,148
/
No. STDR (%)
704 (12.27%)
211 (5.09%)
-7.18
<0.001
No. graded as STDR (%)
771 (13.44%)
274 (6.61%)
-6.83
<0.001
Sensitivity (95% CI)
95.03 (93.42-96.63)
90.05 (86.01-94.09)
-4.98
0.008
Specificity (95% CI)
97.97 (97.58-98.36)
97.87 (97.42-98.32)
-0.10
0.742
PPV (95% CI)
86.77 (84.38-89.16)
69.34 (63.88-74.8)
-17.43
<0.001
NPV (95% CI)
99.3 (99.06-99.53)
99.46 (99.23-99.69)
+0.16
0.318
Accuracy (95% CI)
97.61 (97.22-98.01)
97.47 (96.99-97.95)
-0.14
0.657
HG
No. of patients
5,738
4,263
/
No. STDR (%)
704 (12.27%)
289 (6.78%)
-5.49
<0.001
No. graded as STDR (%)
590 (10.28%)
224 (5.25%)
-5.03
<0.001
Sensitivity (95% CI)
73.72 (70.47-76.97)
57.09 (51.39-62.8)
-16.63
<0.001
Specificity (95% CI)
98.59 (98.26-98.92)
98.52 (98.14-98.89)
-0.07
0.753
PPV (95% CI)
87.97 (85.34-90.59)
73.66 (67.89-79.43)
-14.31
<0.001
NPV (95% CI)
96.41 (95.9-96.91)
96.93 (96.4-97.46)
+0.52
0.169
Accuracy (95% CI)
95.54 (95-96.07)
95.71 (95.1-96.32)
+0.17
0.681
STDR: sight-threatening diabetic retinopathy; PPV: positive predictive value; NPV: negative predictive value; DL: deep learning; HG: trained human graders; CI: confidence interval. value was calculated from chi-squared test for the difference between the first and second screening. value < 0.05.
