Improved Complex Convolutional Neural Network Based on SPIRiT and Dense Connection for Parallel MRI Reconstruction

Algorithm 1

Parallel MRI reconstruction algorithm based on the SPIRiT model and complex convolutional neural network (SPIRiT-Net).

1: Input:

2: Parameters: Total number of training samples , learning rate , decay rate , network parameters , batch size , number of convolution units , and number of complex-valued convolution blocks .

3: fordo

4: Obtain or (in matrix form) by solving the problem (4).

5: 

6: 

7: fordo

8:  ifthen

9:   

10:  else

11:   

12:  end if

13:  fordo

14:   

15:   

16:  end for

17:  

18:  

19: end for

20: ifthen

21:  Calculated loss:

22:  Update parameters:

23: end if

24: end for

25: Output network parameters: