Classification of chest X-ray images by incorporation of medical domain knowledge into operation branch networks

BMC Medical Imaging

Table 3 Pulmonary hypertension classification results from ten-fold cross-validation

Backbone	Model	Accuracy [%]	Sensitivity [%]	Specificity [%]	AUC
ResNet50	Conventional	64.17 ± 5.89	68.56 ± 10.74	59.78 ± 11.48	0.69 ± 0.05
	ABN	63.10 ± 3.65	61.03 ± 7.99	65.17 ± 7.83	0.69 ± 0.04
	OBN1	63.11 ± 4.37	60.52 ± 7.86	65.70 ± 12.91	0.69 ± 0.04
	OBN2	64.95 ± 4.66	64.15 ± 9.21	65.74 ± 15.01	0.70 ± 0.04
	OBN3	64.53 ± 4.03	61.75 ± 9.53	67.31 ± 12.84	0.70 ± 0.04
DenseNet121	Conventional	63.66 ± 4.99	65.60 ± 6.61	61.73 ± 10.30	0.69 ± 0.05
	ABN	61.54 ± 4.94	57.72 ± 9.88	65.35 ± 11.12	0.68 ± 0.06
	OBN1	64.33 ± 4.55	61.77 ± 8.69	66.88 ± 8.50	0.69 ± 0.04
	OBN2	64.05 ± 4.11	57.40 ± 11.06	70.69 ± 5.08	0.69 ± 0.04
	OBN3	64.79 ± 3.59	58.70 ± 7.73	70.88 ± 6.52	0.70 ± 0.04

Bolded numbers indicate the highest score
ResNet50 and DenseNet121 are used as backbone approaches: ABN, attention branch network; OBN1, operation branch network using weight map with a convex hull on mask images of lung field; OBN2, operation branch network using weight maps with a combined mask image of lung field and heart; OBN3, operation branch network using weight maps manually masked to include the heart, produced with a doctor's support

ISSN: 1471-2342