Table 4

Confusion matrices for deep convolutional ensemble and board-certified ophthalmologists showing the mean (and SD) per cent agreement between ‘confident’ predicted labels against the ground-truth labels over the test set

Deep convolutional ensembleOphthalmologists
NormalDRGlaucomaAMDNormalDRGlaucomaAMD
Ground-truth labelsNormal25.1%
(0.9%)		1.3%
(0.9%)		0.0%
(0.0%)		0.0%
(0.0%)		22.1%
(3.0%)		0.3%
(0.6%)		0.2%
(0.6%)		0.2%
(0.5%)
DR5.7%
(1.2%)		18.9%
(1.0%)		0.0%
(0.0%)		0.0%
(0.0%)		7.8%
(5.5%)		12.7%
(6.8%)		0.2%
(0.6%)		2.1%
(1.1%)
Glaucoma5.7%
(2.3%)		0.2%
(0.5%)		18.9%
(2.7%)		0.0%
(0.0%)		4.3%
(4.1%)		0.4%
(1.0%)		18.5%
(4.2%)		2.0%
(0.8%)
AMD2.8%
(0.6%)		1.9%
(1.2%)		0.0%
(0.0%)		19.4%
(1.5%)		0.5%
(0.9%)		2.1%
(1.6%)		0.2%
(0.6%)		26.3%
(2.3%)

  • Green cells indicate agreement between the ground-truth labels and predictions by the deep convolutional ensemble or ophthalmologists, and red cells similarly indicate disagreement

  • AMD, age-related macular degeneration; DR, diabetic retinopathy.