Skip to main content
. Author manuscript; available in PMC: 2018 Dec 1.
Published in final edited form as: J Glaucoma. 2017 Dec;26(12):1086–1094. doi: 10.1097/IJG.0000000000000765

Table 1.

Average AROC values and 95% confidence intervals (CI) across 50 trials for each fully connected layer for each image type.

Comparing classification performance of fully connected layers
fc6 fc7 fc8 fc6+fc7+fc8
Mean 95% CI Mean 95% CI Mean 95% CI Mean 95% CI
RNFL Thickness .973 .972,.974 .962 .961,.964 .956 .954,.957 .973 .972,.974
RGC+ Thickness .938 .937,.940 .936 .934,.937 .932 .931,.933 .935 .934,.936
RNFL Probability .961 .960,.962 .957 .956,.958 .958 .957,.959 .979 .976,.980
RGC+ Probability .920 .920,.921 .920 .920,.920 .913 .911,.914 .950 .949,.951
en face Projection .742 .737,.747 .697 .692,.701 .688 .683,.694 .758 .754,.761
Combined .945 .944,.947 .902 .900,.903 .885 .884,.886 .919 .917,.920