Abstract
The purpose of this article is to elucidate inherent limitations to the performance of parallel MRI. The study focuses on the ultimate signalβtoβnoise ratio (SNR), which refers to the maximum SNR permitted by the electrodynamics of the signal detection process. Using a spherical model object, it is shown that the behavior of the ultimate SNR imposes distinct limits on the acceleration rate in parallel imaging. For low and moderate acceleration, the ultimate SNR performance is nearly optimal, with geometry factors close to 1. However, for high reduction factors beyond a critical value, the ultimate performance deteriorates rapidly, corresponding to exponential growth of the geometry factor. The transition from optimal to deteriorating performance depends on the electrodynamic characteristics of the detected RF fields. In the nearβfield regime, i.e., for low B~0~ and small object size, the critical reduction factor is constant and approximately equal to four for 1D acceleration in the sphere. In the farβfield wave regime the critical reduction factor is larger and increases both with B~0~ and object size. Therefore, it is concluded that parallel techniques hold particular promise for human MR imaging at very high field. Magn Reson Med 52:376β390, 2004. Β© 2004 WileyβLiss, Inc.