Table 8.

Quantitative comparison between the proposed model (SA + ResNet50) and other classification networks (InceptionV3, VGG16, ResNet101, VGG19, and ResNet18 each with or without SA, and also ResNet50 without SA). For all of them we use SGD Optimizer during the training phase.

	TP	TN	FP	FN	Precision	Sensitivity or Recall (%)	F1-Score (%)	Specificity (%)	Accuracy (%)	AUC (%)
SA + ResNet50	127	392	4	5	0.9695	96.21	96.6	99	96.2	98.83
ResNet50	118	386	10	14	0.92	89.4	90.8	96.6	91.7	98.3
SA +VGG16	124	388	8	8	0.939	93.9	93.9	98	93.9	98.87
VGG16	118	385	11	14	0.9147	89.4	90.4	97.2	90.9	98.8
SA + InceptionV3	123	388	8	9	0.93	93.2	93.5	98	93.9	98
InceptionV3	120	384	12	12	0.909	90.9	90.9	97	90.9	97
SA + ResNet101	114	385	11	18	0.912	86.36	97.2	88.7	89.4	97.86
ResNet101	115	384	12	17	0.9055	87.12	97	88	87.9	98.09
SA +VGG19	112	380	16	20	0.875	84.85	96	86.2	85.6	97.6
VGG19	125	391	5	7	0.9615	94.7	98.7	95.4	95.5	99.22
SA + ResNet18	100	378	18	32	0.8475	75.76	80	95.5	80.3	95.32
ResNet18	89	370	26	43	0.7739	67.42	72.1	93.4	72.7	91.41