Table 5.

Comparative average rounds of cross-validation results of five CNN models for TB detection in the combined dataset without augmentation in nine angles

Schema	Dataset	CNN models	Average weighted
			Accuracy (%)	Recall (%)	Precision (%)	F1-Score (%)	Specificity (%)	Kappa value (%)
Without Segmentation Without Augmentation	Montgomery dataset	ResNet50	68.7	73.1	71.3	72.3	61.7	34.9
		Xception	73.9	77.5	77.5	77.5	69.0	46.5
		InceptionRenNetV2	77.5	78.2	85.0	81.4	76.5	53.1
		MobileNet	68.9	73.4	72.5	73.0	62.7	36.2
		EfficientNetB3	82.80	83.3	87.5	85.4	81.5	64.0
	Shenzhen dataset	ResNet50	77.0	76.7	76.7	76.7	77.4	54.1
		Xception	80.1	81.7	76.7	79.1	78.7	60.1
		InceptionRenNetV2	83.8	85.2	81.3	83.2	82.6	67.6
		MobileNet	79.3	79.4	78.2	78.8	79.2	58.6
		EfficientNetB3	87.6	89.4	85.0	87.1	86.1	75.2
	Combination dataset	ResNet50	88.7	82.8	87.8	85.2	92.6	76.1
		Xception	87.4	79.6	88.7	83.9	92.9	73.7
		InceptionRenNetV2	90.6	84.7	91.1	87.8	94.6	80.2
		MobileNet	88.6	81.3	89.9	85.4	93.7	76.1
		EfficientNetB3	90.3	84.5	90.1	87.2	94.0	79.4
With Segmentation Without Augmentation	Montgomery dataset	ResNet50	73.2	77.2	76.3	76.7	67.8	45.1
		Xception	77.0	79.3	81.2	73.2	73.2	52.2
		InceptionRenNetV2	81.2	83.8	83.8	83.8	77.6	61.3
		MobileNet	72.5	76.9	75.0	75.9	66.7	43.8
		EfficientNetB3	85.5	88.5	86.3	87.3	81.7	70.4
	Shenzhen dataset	ResNet50	84.7	86.6	81.6	84.0	83.1	69.5
		Xception	86.1	88.0	85.5	84.5	84.5	72.2
		InceptionRenNetV2	89.6	90.8	87.7	89.2	88.5	79.1
		MobileNet	86.3	88.3	83.1	85.6	84.5	72.5
		EfficientNetB3	90.3	91.7	88.3	90.0	89.1	80.6
	Combination dataset	ResNet50	88.5	84.3	84.7	84.5	91.0	74.4
		Xception	89.9	86.1	86.7	86.4	91.2	78.3
		InceptionRenNetV2	90.8	87.5	87.8	87.5	92.8	80.4
		MobileNet	90.7	86.9	88.2	87.5	93.0	80.1
		EfficientNetB3	91.2	87.4	88.9	88.2	93.4	81.1

The best result of each set has been bolded