Skip to main content
. Author manuscript; available in PMC: 2020 Nov 1.
Published in final edited form as: Ophthalmology. 2019 Jun 11;126(11):1533–1540. doi: 10.1016/j.ophtha.2019.06.005

Table 2.

Performance of the three deep learning models: (i) geographic atrophy model (‘GA model’), (ii) central geographic atrophy model (‘cGA’ model), and (iii) centrality detector model.

GA detection cGA detection from all eyes cGA detection from eyes with GA
GA model Retinal specialists cGA model Retinal specialists Centrality detector model Retinal specialists
(95% CI) (95% CI) (95% CI) (95% CI) (95% CI) (95% CI)
Accuracy 0.965 (0.959, 0.971) 0.975 (0.971, 0.980) 0.966 (0.957, 0.975) 0.990 (0.987, 0.993) 0.762 (0.725, 0.799) 0.735 (0.445, 1.000)
Kappa 0.611 (0.533, 0.689) 0.438 (0.304, 0.571) 0.501 (0.451, 0.552) 0.344 (0.156, 0.532) 0.513 (0.442, 0.584) 0.533 (0.176, 0.891)
Sensitivity 0.692 (0.560, 0.825) 0.588 (0.468, 0.707) 0.763 (0.641, 0.885) 0.448 (0.255, 0.641) 0.782 (0.618, 0.945) 0.878 (0.722, 1.000)
Specificity 0.978 (0.970, 0.985) 0.982 (0.978, 0.985) 0.971 (0.960, 0.982) 0.993 (0.989, 0.996) 0.729 (0.543, 0.916) 0.703 (0.332, 1.000)
Precision 0.584 (0.491, 0.676) 0.368 (0.230, 0.505) 0.394 (0.341, 0.448) 0.296 (0.115, 0.477) 0.799 (0.710, 0.888) 0.626 (0.273, 0.979)