Exudate Detection for Diabetic Retinopathy Using Pretrained Convolutional Neural Networks

Table 1

Comparative classification accuracy results of the proposed model with individual CNN models for exudate detection using e-Ophtha dataset.

CNN models	Data splitting
	Training (%)	Testing (%)	F1 score	Recall	Precision	Classification accuracy (%)
Inception-v3	70	30	0.95	0.98	0.92	92.50%
	80	20	0.93	0.94	0.92	92.90%
	90	10	0.94	0.92	0.96	93.67%
ResNet-50	70	30	0.94	0.98	0.91	90.67%
	80	20	0.94	0.98	0.90	95.70%
	90	10	0.98	0.97	0.99	97.80%
VGG-19	70	30	0.94	0.99	0.90	92.33%
	80	20	0.94	0.93	0.95	95.80%
	90	10	0.94	0.90	0.89	93.50%
Proposed model	70	30	0.96	0.96	0.97	97.98%
	80	20	0.96	0.97	0.95	98.43%
	90	10	0.95	0.96	0.95	97.90%