Table 21.

The performance of DAG-based networks based on augmentation and without augmentation using original & pre-processed DR images of Combined Dataset

(a) Using Original Combined Dataset images
Network Name	Confusion Matrix		Without Augmentation					Confusion Matrix		After Augmentation
			ACC %	Sen	Sp	Pr	F1			ACC %	Sen	Sp	Pr	F1
inceptionv3	325	125	69.2	0.66	0.72	0.70	0.68	415	35	89.55	0.87	0.92	0.92	0.89
	152	298						59	391
densenet201	339	111	69.6	0.64	0.75	0.72	0.68	414	36	89.11	0.86	0.92	0.92	0.89
	162	288						62	388
Resnet50	339	111	71.7	0.68	0.75	0.73	0.71	414	36	90.22	0.88	0.92	0.92	0.90
	143	307						52	398
Resnet101	348	102	0.74	0.71	0.77	0.76	0.73	426	24	93.55	0.92	0.95	0.95	0.93
	132	318						34	416
xception	339	111	70.3	0.65	0.75	0.73	0.69	417	33	91.33	0.90	0.93	0.92	0.91
	156	294						45	405
inceptionresnetv2	328	122	67.8	0.63	0.73	0.70	0.66	406	44	88.33	0.86	0.90	0.90	0.88
	167	283						61	389
nasnetlarge	330	120	69.3	0.65	0.73	0.71	0.68	416	34	90.11	0.88	0.92	0.92	0.90
	156	294						55	395
Using Pre-processed Combined Dataset Images
Network Name	Confusion Matrix		ACC %	Sen	Sp	Pr	F1	Confusion Matrix		ACC %	Sen	Sp	Pr	F1
inceptionv3	339	111	72	0.69	0.75	0.74	0.71	415	35	90.88	0.89	0.93	0.93	0.88
	141	309						72	378
densenet201	343	107	72.3	0.68	0.76	0.74	0.71	418	32	91	0.89	0.93	0.93	0.89
	142	308						66	384
Resnet50	352	98	73.7	0.69	0.78	0.76	0.73	420	30	91.44	0.90	0.93	0.93	0.91
	138	312						50	400
Resnet101	358	92	75.6	0.72	0.80	0.78	0.75	440	10	97.33	0.97	0.98	0.98	0.97
	127	323						14	436
xception	340	110	71.3	0.67	0.76	0.73	0.70	414	36	92.44	0.91	0.94	0.93	0.88
	148	302						66	384
inceptionresnetv2	329	121	69	0.65	0.73	0.71	0.68	402	48	89.77	0.89	0.91	0.91	0.87
	158	292						70	380
nasnetlarge	328	122	70.6	0.68	0.73	0.72	0.70	410	40	91.44	0.89	0.94	0.94	0.88
	142	308						68	382