Comparing code-free deep learning models to expert-designed models for detecting retinal diseases from optical coherence tomography

Table 1 Performance and evaluation of the CFDL video model

	Total	TP	FP	TN	FN	AUPRC	PPV	SN	ACC
Overall	118	NR	NR	NR	NR	0.984	94.1%	94.1%	94.1%
Normal	52	50	2	64	2	0.993	96.2%	96.2%	96.2%
MH	12	11	0	106	1	1.000	100.0%	91.7%	91.7%
ERM	21	19	3	94	2	0.954	86.4%	90.5%	90.5%
AMD	19	18	0	99	1	1.000	100.0%	94.7%	94.7%
DME	14	13	2	102	1	0.957	86.7%	92.9%	92.9%

TP, true positives; FP, false positives; TN, true negatives; FN, false negatives; AUPRC, area under the precision-recall curve; SN, sensitivity (recall); PPV, positive predictive value (precision); ACC, accuracy; MH, macular hole; ERM, epiretinal membrane; AMD, age-related macular degeneration; DME, diabetic macular oedema

ISSN: 2056-9920