Automated cardiovascular magnetic resonance image analysis with fully convolutional networks

Table 1 The network architecture

Scale	Size	Convolution
1	192×192	3×3, 16
		3×3, 16
2	96×96	3×3, 32
		3×3,32
3	48×48	3×3, 64
		3×3, 64
		3×3, 64
4	24×24	3×3, 128
		3×3, 128
		3×3, 128
5	12×12	3×3, 256
		3×3, 256
		3×3, 256
upsample and concatenate scale 1 to 5 features
predict	192×192	1×1, 64
		1×1, 64
		1×1, K

The first two columns list the resolution scale and feature map size. The third column lists the convolutional layer parameters, with “ 3×3,16” denoting 3×3 kernel and 16 output features. The last convolutional layer outputs K features, with K denoting the number of label classes

ISSN: 1532-429X