Automated Segmentation of Colorectal Tumor in 3D MRI Using 3D Multiscale Densely Connected Convolutional Neural Network

Table 1

Quantitative comparison of colorectal tumor segmentation results.

Methods	Performance metrics
	DSC	RR	ASD (mm)
3D FCNNs [15]	0.6519 ± 0.0181	0.6858 ± 0.1017	4.2613 ± 3.1603
3D U-net [12]	0.7227 ± 0.0128	0.7463 ± 0.0302	3.0173 ± 3.0133
DenseVoxNet [16]	0.7826 ± 0.0146	0.8061 ± 0.0187	2.7253 ± 2.9024
3D MSDenseNet (proposed method)	0.8406 ± 0.0191	0.8513 ± 0.0201	2.6407 ± 2.7975
3D FCNNs + 3D level set [15]	0.7591 ± 0.0169	0.7903 ± 0.0183	3.0029 ± 2.9819
3D U-net + 3D level set	0.8217 ± 0.0173	0.8394 ± 0.0193	2.8815 ± 2.6901
DenseVoxNet + 3D level set	0.8261 ± 0.0139	0.8407 ± 0.0177	2.5249 ± 2.8004
3D MSDenseNet + 3D level set (proposed method)	0.8585 ± 0.0184	0.8719 ± 0.0195	2.5401 ± 2.402