Peng Lai1, Michael Lustig2,3, Anja CS. Brau1, Shreyas Vasanawala4, Philip J. Beatty1, Marcus Alley2
1Applied Science Laboratory, GE Healthcare, Menlo Park, CA, United States; 2Electrical Engineering, Stanford University, Stanford, CA, United States; 3Electrical Engineering and Computer Science, University of California, Berkeley, CA, United States; 4Radiology, Stanford University, Stanford, CA, United States
Conventional L1SPIRiT reconstruction enables highly-accelerated MRI by combining parallel imaging and compressed sensing but suffers from impractically long reconstruction time. This work developed a new efficient L1SPIRiT algorithm (ESPIRiT) to address the computation challenge from three perspectives: 1. reducing the computation complexity based on Eigenvector calculations, 2. reducing the number of pixels to process based on pixel-specific convergence, 3. reducing the number of iterations using parallel imaging initialization. ESPIRiT was compared with L1SPIRiT on in-vivo datasets. Our results show that ESPIRiT can improve image quality and reconstruction accuracy with >10 faster computation compared to L1SPIRiT.