Abstract
The prolonged time required for Magnetic Resonance Imaging (MRI) procedures often causes stress and anxiety in patients. Patients' elongated exposure to a narrow bore of a scanner increases healthcare costs. Additionally, long scan times increase patient discomfort and cause motion artifacts in images. As a clinical MRI method, Generalized Autocalibrating Partially Parallel Acquisition (GRAPPA) is a parallel imaging technique that shortens data acquisition time and increases spatial resolution by employing multiple phased-array coils. Missing data are reconstructed by solving a linear time-invariant (LTI) system. When a large size of an interpolation kernel is used, the number of equations of the LTI system is significantly increased and the reconstruction time is delayed. Both software-based and hardware-based approaches have been proposed to accelerate the GRAPPA reconstruction. Dimension reduction is often used as the software-based approach which generates virtual coils and reduces the actual number of physical coils, so the total number of LTI equations is decreased. Field-programmable gate array (FPGA) was employed as a hardware-based approach to solve the LTI system. In this research, a quantum computing method was proposed to accelerate GRAPPA reconstruction. Different from the dimension reduction and FPGA techniques, quantum computing is an emerging computing paradigm that can accelerate computing speed with quantum bits (qubits) rather than classical bits. We decomposed GRAPPA reconstruction into two components: calibration and synthesis, and then applied the D-Wave quantum computing service to accelerate the calibration process. A few calibration coefficients are estimated through solving an LTI system on the D-Wave quantum computer. Then, those calibrated coefficients are used to synthesize missing data in under sampled k-space to meet the requirements of the Nyquist sampling rate and avoid aliasing artifacts. Experimental results show that quantum computing can successfully solve the calibration problem in GRAPPA MRI. The calibrated coefficients and reconstruction results are the same as the outcomes completed on a local computer which has the classical von Neumann architecture. Quantum computing speeds have been quantitatively analyzed and it is promising to accelerate GRAPPA. One drawback of the current implementation is that the D-Wave quantum computing service cannot accept over 30 variables of an LTI system, and this may be caused by the limited qubits which can be used on the quantum computing service now.