Automatic lesion segmentation using atrous convolutional deep neural networks in dermoscopic skin cancer images

BMC Medical Imaging

Table 4 Performance comparison of the DilatedSkinNet with its modified architectures

Methods	ISIC 2016				ISIC 2017				ISIC 2018
	ACC	JAC	DICE	Time (s)	ACC	JAC	DICE	Time (s)	ACC	JAC	DICE	Time (s)
DilatedSkinNet	0.940	0.887	0.940	10	0.879	0.817	0.874	9	0.942	0.891	0.942	14
Without atrous	0.853	0.735	0.513	12	0.758	0.556	0.460	16	0.825	0.635	0.510	21
Without leakyReLU	0.935	0.855	0.622	11	0.874	0.784	0.578	14	0.863	0.736	0.556	19
Without augmentation	0.945	0.884	0.679	11.8	0.885	0.788	0.559	12.6	0.942	0.854	0.643	16.5
Sigmoid layer	0.450	0.329	0.496	27	0.465	0.338	0.506	27	0.499	0.331	0.498	51

ISSN: 1471-2342