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

Table 5

Results of all systems in this work for diagnosing endoscopy images of the gastroenterology dataset.

Methodology	Classes	Dyed-lifted-polyps	Dyed-resection-margins	Esophagitis	Normal-cecum	Normal-pylorus	Normal-z-line	Polyps	Ulcerative-colitis	Accuracy (%)
Hybrid method	VGG-16 + SVM	97	99	96.5	95.5	96.5	97.5	96	93.5	96.4
	DenseNet-121 + SVM	92.5	96.5	95.5	95	97	98.5	95.5	94	95.6
Fusing features before PCA	VGG-16 + DenseNet-121	97.5	99	98.5	98.5	97.5	96.5	97.5	97.5	97.8
Fusing features after PCA	VGG-16 + DenseNet-121	95.5	99	98.5	96.5	97.5	98.5	97.5	95.5	97.3
Fusion features	VGG-16 and handcrafted	98	99.5	98.5	98.5	98.5	99.5	98.5	97.5	98.6
	DenseNet-121 and handcrafted	98	99.5	98	100	99.5	99.5	98.5	98.5	98.9