Figure 5.

Temporal data processing pipeline with neural network parameter inference. The letters follow the same order as Fig. 3, where the neural network architecture corresponding to the multi-pathway network is shown in (f). After dimensionality reduction via SVD subspace projection (c), gridding, FFT, coil estimation and combination (c–e) are all linear operations that do not affect the input of the compressed signal into the neural network. The output of the network is subsequently used to synthesize image contrasts (g).