Privacy-preserving Outsourcing of Parallel Magnetic Resonance Image Reconstruction

With the rapid technological advances in parallel imaging reconstruction, Magnetic Resonance Imaging (MRI) has been increasingly popularized for clinical diagnosis. Among the state-of-the-art approaches, the Simultaneous Auto-calibrating and k-space Estimation (SAKE) realizes a calibration-free reconstruction with high-quality results. However, its reconstruction procedures are still time-consuming under clinical settings in terms of the large data matrix. To accelerate the reconstruction speed in SAKE, a novel idea is outsourcing the complex computation to the cloud and leverage the abundant computing resources. But how to concurrently protect the patients' data privacy becomes the major challenge. To address this issue, we propose to apply a series of random matrices and transformations to hide the original data matrix, which enables the clinic to securely derive the approximate solution and check its correctness. The efficiency of the proposed protocol is then validated in the experiment compared to original SAKE.