Table 2.

Validation of low quality and high quality ResNet models trained on 8 $\times$ 8 mm images against 8 $\times$ 8 mm and 6 $\times$ 6 mm OCTA superficial slab images shows robust quality assessment.

	8 $\times$ 8 mm validation set (n = 69)	6 $\times$ 6 mm validation set (n = 32)
Low quality ResNet (trained on low quality 8 $\times$ 8 mm image cutoff)	AUC = 0.99, 95% CI [0.98–1.00] Accuracy = 95.3% Cohen’s Kappa = 0.90	AUC = 0.83, 95% CI [0.69–0.98] Accuracy = 87.5% Cohen’s Kappa = 0.7
Low quality CNN (trained on low quality 8 $\times$ 8 mm image cutoff)	AUC = 0.97, 95% CI [0.95–0.98] Accuracy = 91 % Cohen’s Kappa = 0.82	AUC = 0.79, 95% CI [0.63–0.96] Accuracy = 68.8% Cohen’s Kappa = 38.2
High quality ResNet (trained on high quality 8 $\times$ 8 mm image cutoff)	AUC = 0.97, 95% CI [0.96–0.99] Accuracy = 93.5% Cohen’s Kappa = 0.81	AUC = 0.85, 95% CI [0.55–1.00] Accuracy = 71.9% Cohen’s Kappa = 0.41
High quality CNN (trained on high quality 8 $\times$ 8 mm image cutoff)	AUC = 0.94, 95% CI [0.91–0.97] Accuracy = 90.1% Cohen’s Kappa = 0.71	AUC = 0.67, 95% CI [0.19–1.00] Accuracy = 71.8% Cohen’s Kappa = 0.41

AUC area under the curve.