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

BMC Medical Imaging

Table 2 Teikyo dataset classification results from ten-fold cross-validation

Backbone	Model	Accuracy [%]	Sensitivity [%]	Specificity [%]	AUC
ResNet50	Conventional	93.16 ± 1.49	89.51 ± 3.03	96.80 ± 1.19	0.97 ± 0.01
	ABN	93.54 ± 1.55	91.17 ± 2.86	95.90 ± 1.43	0.97 ± 0.01
	OBN1	93.62 ± 1.48	90.29 ± 3.19	96.95 ± 1.91	0.98 ± 0.01
	OBN2	93.44 ± 1.81	90.78 ± 2.89	96.11 ± 2.24	0.98 ± 0.01
	OBN3	93.23 ± 1.53	89.90 ± 3.23	96.55 ± 1.78	0.98 ± 0.01
DenseNet121	Conventional	93.99 ± 1.22	91.07 ± 2.29	96.90 ± 0.89	0.98 ± 0.01
	ABN	93.59 ± 1.45	90.68 ± 2.05	96.50 ± 2.07	0.98 ± 0.01
	OBN1	93.50 ± 1.51	90.19 ± 2.80	96.80 ± 1.20	0.98 ± 0.01
	OBN2	93.64 ± 1.26	90.78 ± 3.08	96.50 ± 1.21	0.98 ± 0.01
	OBN3	93.62 ± 2.03	90.19 ± 3.70	97.05 ± 1.38	0.98 ± 0.01

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