Table 4 Effect of image-related factors specifically compression levels.

From: Technical and imaging factors influencing performance of deep learning systems for diabetic retinopathy

   

Compression level – image file size

   

350 KB

300 KB

250 KB

200 KB

150 KB

SiDRP

Value (95% CI)

AUC

0.936 (0.927–0.944)

0.921 (0.908–0.932)

0.900 (0.885–0.913)

0.896 (0.881–0.910)

0.891 (0.876–0.905)

P value for AUC comparison

Reference

0.261

<0.001

<0.001

<0.001

Sensitivity

90.5% (87.3–93.1%)

85.9% (82.2–89.0%)

83.5% (79.7–86.9%)

85.6% (81.9–88.8%)

90.5% (87.3–93.1%)

Specificity

91.9% (91.6–92.2%)

92.5% (92.3–92.8%)

88.8% (88.5–89.2%)

85.3% (84.9–85.7%)

72.4% (71.9–72.8%)

  1. P value was calculated by bootstrap method, taking 350 KB as the reference for comparison against.
  2. Dataset used for evaluation of different compression levels is Singapore integrated Diabetic Retinopathy Programme (SiDRP) 2014 to 2015.
  3. AUC area under receiver operating curve, CI confidence interval, KB kilobytes, SiDRP Singapore integrated Diabetic Retinopathy Programme.