Abstract. Real-time parallel MRI reconstruction was demonstrated using a hybrid implementation of the TGRAPPA algorithm. The GRAPPA coefficients were calculated in k-space and applied in the image domain after appropriate transformation, thereby achieving improved speed and excellent image quality. Adaptive B1-weighted combining of the per coil images permitted use of pre-calculated composite image domain weights providing significant decrease in computation. The weight calculation was decoupled from the real-time image reconstruction as a parallel processing thread which was updated in an adaptive manner to speed convergence in the event of interactive change in scan plane. The computation was parallelized and implemented on a general purpose multicore architecture. Reconstruction speeds of 65-70 frames per second were achieved with a matrix of 192x144 with 15 coils.
Haris Saybasili, Peter Kellman, J. Andrew Derbys