MDU-Net: A Convolutional Network for Clavicle and Rib Segmentation from a Chest Radiograph

Table 2

Performance comparison of the single-task segmentation networks.

	Evaluation	FCN-8s [30]		Deeplabv3+ [32]		U-Net [31]		MDU_Net
		Mean	Std	Mean	Std	Mean	Std	Mean	Std
Clavicle	DSC	90.51	3.87	90.28	4.97	92.53	5.42	93.78	2.21
	Precision	92.16	3.10	91.59	4.15	93.93	3.09	94.53	3.17
	Recall	89.26	6.14	89.47	8.05	91.59	7.45	93.03	3.53
	Jaccard	82.66	5.56	82.28	7.58	86.10	7.21	88.29	3.98
Anterior ribs	DSC	78.41	4.77	75.76	5.37	79.59	5.26	80.95	5.38
	Precision	83.1	10.04	77.82	10.18	83.43	10.4	83.18	11.02
	Recall	75.74	7.89	75.41	8.27	78.37	8.13	81.25	7.49
	Jaccard	64.49	6.25	60.98	6.8	66.10	6.88	68.09	7.2
Posterior ribs	DSC	85.39	2.67	84.66	3.44	87.48	2.83	89.06	2.45
	Precision	87.59	3.91	85.33	4.64	89.56	4.13	89.69	4.22
	Recall	83.54	4.3	84.28	4.94	85.73	4.32	88.67	3.62
	Jaccard	74.50	4.01	73.40	5.11	77.75	4.44	80.28	3.98
All	DSC	85.62	2.81	84.24	3.06	86.47	2.96	88.38	2.94
	Precision	88.82	4.34	86.2	5.1	89.89	4.56	90.39	4.71
	Recall	83.03	5.41	82.91	6.39	83.66	5.46	86.72	5.79
	Jaccard	74.86	4.21	72.77	5.08	76.16	4.56	79.18	4.68