Table 1.

Encoding layers of the proposed late fusion U-net. The encoder is composed of 5 convolution blocks (Conv Block), maxpooling operations (2D MaxPooling) and dropout. The Conv Block is made of: 2D convolution (3*3)+ batch normalization + 2D convolution (3*3)+ batch normalization.

Layer (type)	Output shape
Conv Block 1	192*192*8
2D MaxPooling	96*96*8
Conv Block 2	96*96*16
2D MaxPooling	48*48*16
Conv Block 3	48*48*32
2D MaxPooling	24*24*32
Conv Block 4	24*24*64
Dropout + 2D Maxpooling	12*12*64
Conv Block 5 + Dropout	12*12*128
Concatenation	12*12*640