The Optimization Parameters Selection And Weight Adjustment Of GRAPPA For Parallel MRI

Magnetic resonance imaging（MRI） is a non-invasive radiological imaging technique,compared with the X-ray or computed tomography, the biggest advantages of MRI are itssafety speed and accuracy, and do no harm to the human body. However, due to the longerscan time of conventional magnetic resonance imaging, imaging speed is slower, and can’tmeet the requirements of real-time imaging of the heart and other high-end clinicalapplications. The parallel magnetic resonance imaging broke through the traditionalmagnetic resonance imaging by the RF hardware and the limit of the magnetic field gradientperformance. Among them, Generalized autocalibrating partially parallel acquisitions(GRAPPA) is widely used in the clinical routine. However, there is no detailed analysis onthe relationship of the image quality on dependency of the reconstruction and acquisitionparameters for32-channel coil, but also no relevant report about apply neural networks tothe GRAPPA algorithm. In this work, we study the relationship between reconstruction andacquisition parameters of GRAPPA and image quality in detail. And apply neural network tothe GRAPPA algorithm. The main works are described as follows:Firstly, we study of the dependencies between the GRAPPA reconstruction algorithmbased on the32-channel coil in sagittal and transverse brain data acquisition andreconstruction parameters and image quality in detail. The results show that the choice ofoptimal image reconstruction parameters of kernel size limited by b_x=5-7, and b_y=2-4, whilethe choice of the number of autocalibration signal lines must be greater than the lower limitN_acs=10-14.Secondly, neural networks have been applied to GRPPA algorithm, theories and methodsof the GRAPPA algorithm based on neural networks have been proposed, and a preliminaryexploration and research has been made, which opened a new road for future study. Theresults show that with R=2, the reconstructed image quality is best, but as R increases, theimage artifacts are more serious. Meanwhile, the reconstructed image quality is also affectedby the number of autocalibration signal lines. The more the number of autocalibration signallines, the better the image quality.