Trajectory Optimization and Machine Learning Radiofrequency Pulses for Enhanced Magnetic Resonance Imaging

Abstract

High field magnetic resonance imaging (MRI) offers several advantages over imaging at low field strengths, namely increased spectral resolution, better contrast due to longer T1 relaxation, higher signal to noise ratio (SNR), and better parallel imaging performance. However, many imaging techniques require strong flip angle uniformity and fast readouts, which are susceptible to trajectory errors. Optimization and machine learning methods are introduced to reduce image artifacts and decrease RF inhomogeneities in high field acquisitions. This is accomplished by employing algorithms that 1) exploit redundancies inherent in parallel imaging and 2) exploit redundant information in multi-subject data to learn characteristic relationships between RF and image parameters. First, an algorithm to reduce trajectory errors--Trajectory Auto-Corrected image Reconstruction (TrACR)-- is presented. TrACR was evaluated with in vivo 7 Tesla (7T) brain data from non-Cartesian acquisitions. TrACR reconstructions reduced blurring and streaking artifacts and bear similar quality to images reconstructed using trajectory measurements. Second, an extension of TrACR is presented for echo planar imaging acquisitions to reduce trajectory and phase errors. EPI-TrACR is validated in vivo at 7T, at multiple acceleration and multishot factors, and in a time series, and consistently reduces image artifacts. Finally, to improve transmit field uniformity, a method is introduced for predicting tailored RF shims. RF-shim Prediction by Iteratively Projected Ridge Regression (PIPRR) was validated in simulation for single-slice shimming for 100 phantom human heads. PIPPR-predicted shims reduced profile inhomogeneity and maintained comparable specific absorption rate (SAR) efficiency and homogeneity to that of directly designed shims. PIPRR predictions for a new patient require just milliseconds, reducing compute time for RF shimming by orders of magnitude.