Wireless capsule endoscopy multiclass classification using three-dimensional deep convolutional neural network model

Table 4 Evaluation metrics of multiclass classification for different deep neural network models: frame-based scheme

		3D	2D	AlexNet	Inception-V3	ResNet-18	SqueezeNet	DenseNet-201
Sensitivity (%)	Lesion	94.52	91.94	75.38	77.74	77.58	81.13	85.48
	Normal	99.50	86.94	97.72	96.89	97.74	97.78	99.10
	Poor	99.76	98.89	98.51	98.78	98.63	98.93	98.10
Specificity (%)	Lesion	99.76	93.15	98.62	98.23	98.75	98.70	99.22
Specificity (%)	Normal	98.92	98.07	94.20	95.51	95.06	95.85	95.85
	Poor	99.62	98.78	99.14	98.71	98.90	99.22	99.84
PPV (%)	Lesion	98.16	64.12	88.61	85.46	89.24	89.34	93.64
	Normal	98.88	97.73	93.92	95.40	95.01	95.77	95.82
	Poor	99.42	98.13	98.70	98.02	98.30	98.78	99.75
NPV (%)	Lesion	99.27	98.86	96.57	97.05	97.05	97.50	98.07
	Normal	99.51	88.69	97.83	96.96	97.77	97.82	99.11
	Poor	99.84	99.28	99.02	99.21	99.12	99.31	98.79
F₁ score	Lesion	98.71	77.79	92.42	90.89	92.98	93.24	95.81
	Normal	99.19	92.99	95.83	96.18	96.37	96.79	97.44
	Poor	99.63	98.70	98.86	98.61	98.71	99.05	99.27
Accuracy (%)	Lesion	99.15	93.00	95.73	95.81	96.24	96.62	97.60
	Normal	99.20	92.63	95.88	96.19	96.38	96.80	97.44
	Poor	99.68	98.82	98.89	98.74	98.80	99.10	99.16

ISSN: 1475-925X