Hybrid Techniques for Diagnosing Endoscopy Images for Early Detection of Gastrointestinal Disease Based on Fusion Features

Table 3

ANN performance by combining VGG-16 and ResNet-18 features before and after PCA.

System	Disease types	Accuracy (%)	Sensitivity (%)	Precision (%)	Specificity (%)	AUC (%)
ANN with combining CNN before PCA	Dyed_lifted_polyps	97.50	97.30	98.50	99.45	99.10
	Dyed_resection_margins	99.00	99.25	98.50	99.60	98.56
	Esophagitis	98.50	98.40	97.50	99.50	98.70
	Normal_cecum	98.50	97.10	98.50	99.70	99.54
	Normal_pylorus	97.50	96.35	97.50	99.68	99.42
	Normal_z_line	96.50	95.80	95.50	99.75	98.65
	Polyps	97.50	96.75	96.10	98.80	97.85
	Ulcerative_colitis	97.50	97.12	97.50	99.64	99.20
	Average ratio	97.80	97.26	97.45	99.52	98.88
ANN with combining CNN after PCA	Dyed_lifted_polyps	95.5	95.22	97.90	99.50	99.40
	Dyed_resection_margins	99	98.85	97.50	100	98.95
	Esophagitis	98.5	97.95	95.20	98.85	98.50
	Normal_cecum	96.5	96.32	98.50	99.65	99.20
	Normal_pylorus	97.5	97.28	99.00	99.60	99.62
	Normal_z_line	98.5	97.75	97.50	99.45	99.57
	Polyps	97.5	97.35	95.60	98.75	97.86
	Ulcerative_colitis	96.50	94.86	97.40	100	99.30
	Average ratio	97.30	95.49	97.33	99.48	99.05