HTGRAPPA: TGRAPPA BASED B1-WEIGHTED IMAGE DOMAIN RECONSTRUCTION FOR REAL-TIME MRI

Introduction: TSENSE [1] based real-time reconstruction software for interventional applications has been previously described [2]. In this work, we present a parallel imaging algorithm based on TGRAPPA [3] for real-time MRI, called HTGRAPPA and its real-time, low latency implementation suitable for interventional MR applications. Our method calculates GRAPPA coefficients in k-space, but applies them in the image domain to avoid time-consuming convolution operations [4] allowing reconstruction fast enough for real-time imaging. In HTGRAPPA, image domain GRAPPA weights were combined into composite unmixing coefficients using adaptive B1-map estimates and optimal noise weighting to eliminate per coil reconstructions. That makes it possible to reconstruct images in the image domain by pixel-by-pixel multiplication and summing, instead of time-consuming convolution operations in k-space. Weight-sets were computed asynchronously to the image reconstruction, and updated quickly to adapt to changes in the image plane and coil sensitivity profiles. Our algorithm provides constant reconstruction performance, independent of the acceleration rate and GRAPPA kernel size. We evaluated our method using 30-coil, rate 4 dataset and compared it to TGRAPPA and TSENSE in performance and image quality. HTGRAPPA reconstruction algorithm was up to 265 times faster than TGRAPPA with no reduction in image quality. A frame rate exceeding 70 was reached on previously acquired data, which is more than sufficient for real-time data rates. Additionally, HTGRAPPA doesn’t exhibit pre-folding artifacts when small FOV is used.