Glaucoma Detection Using Image Processing and Supervised Learning for Classification

Table 3

Comparison of performance of proposed method for PSGIMSR with benchmark data sets.

Data set	Technique used	Accuracy	Precision	Sensitivity	Specificity	F1 score
PSGIMSR (D1)	GoogleNet	86.86	90.40	81.44	91.93	0.8568
	VGG	87.04	89.80	82.08	91.54	0.8576
	ResNet	88.60	91.60	83.72	93.03	0.8748
	Proposed algorithm	91.11	94.18	86.55	95.20	0.9021
DRIONS-DB (D2)	GoogleNet	90.90	90.80	89.37	92.22	0.9007
	VGG	91.63	90.40	91.12	92.05	0.9076
	ResNet	92.72	92.40	91.67	93.62	0.9203
	Proposed algorithm	95.63	95.60	94.84	96.30	0.9521
HRF (D3)	GoogleNet	95.33	94.67	95.94	94.73	0.9530
	VGG	96.00	97.33	94.80	97.26	0.9605
	ResNet	96.67	96.00	97.29	96.05	0.9664
	Proposed algorithm	98.67	97.33	100	97.40	0.9864
Data set	Technique used	Accuracy	Precision	Sensitivity	Specificity	F1 score
DRISHTI-GS (D4)	GoogleNet	92.07	89.03	85.71	95.05	0.8734
	VGG	91.08	87.09	84.37	94.20	0.8571
	ResNet	93.06	90.96	87.03	95.91	0.8895
	Proposed algorithm	95.64	96.12	90.30	98.23	0.9312
Combined (D5)	GoogleNet	83.40	87.50	75	90.69	0.8076
	VGG	83.73	86.87	75.81	90.38	0.8097
	ResNet	85.56	89.16	77.81	92.06	0.8310
	Proposed algorithm	88.96	93.95	81.26	95.53	0.8714

Bold represents the proposed algorithm values.