Fig. 2.

(a) Illustration of the proposed Dense-Unet for T2WI reconstruction with T1WI and under-sampled T2WI as concatenated input; (b) illustration of the detailed configuration of the dense block. Note that we implement the input in (a) as six consecutive axial slices (with three from fully-sampled T1WI and three from under-sampled T2WI). In (b), each dense block consists of five convolutional layers. The growth rate is set to 16, and the output of each block has 80 feature maps.