Multimodal Retinal Image Analysis via Deep Learning for the Diagnosis of Intermediate Dry Age-Related Macular Degeneration: A Feasibility Study

Table 1

The accuracy, sensitivity, and specificity of each CNN design (single modality vs. dual modality vs. multimodality) in classification of three cohorts of YH, OH, and AMD.

	Patient cohort
	YH	OH	AMD
Single modality CNN	Overall accuracy 94.4%
OCT sensitivity (%)	99.6	98.9	77.8
OCT specificity (%)	98.8	86.7	100
Single modality CNN	Overall accuracy 91.9%
OCT-A sensitivity (%)	95.5	83.2	97.6
OCT-A specificity (%)	99.6	96.2	76.4
Single modality CNN	Overall accuracy 93.8%
CFP sensitivity (%)	81.0	84.6	100
CFP specificity (%)	94.4	78.6	86.7
Dual modality CNN (OCT + OCT-A)	Overall accuracy 96.7%
Sensitivity (%)	100	95.7	92.1
Specificity (%)	97.6	94.6	98.7
Dual modality CNN (OCT + CFP)	Overall accuracy 92.9%
Sensitivity (%)	98.1	96.3	100
Specificity (%)	100	97.0	96.0
Multimodality CNN (OCT + OCT-A + CFP)	Overall accuracy 99.8%
Sensitivity (%)	100	99.3	100
Specificity (%)	100	100	99.2