Faster MRI through portable Resonance Gradient Coils with seamless calibration and reconstruction

About This Grant

PROJECT SUMMARY/ABSTRACT MRI is limited by slow encoding. This has resulted in undesirably lengthy MR exams, and long scheduling time that hampers timely diagnosis. The slow encoding also limits the spatial and temporal resolutions of in vivo scanning, which limits our ability to extract key structural and physiological information in numerous clinical and neuroscientific applications. The main thread of technology development for faster imaging has been in array reception and constrained reconstructions, which has provided impressive speed gains. Nonetheless, significant further gain is highly desirable, with complementary research into advanced gradient coil design being actively pursued, via head-insert, non-linear arrays, and resonance designs. These developments have resulted in exciting speed gains, but their wide-spread usage is hampered by: i) expensive/specialized hardware that requires complex calibration, and ii) computationally intensive image reconstruction; limiting their impact. The goal of this proposal is to overcome these constrains by developing portable resonance gradient coils (RGC) that are low-cost and simple to calibrate, and to develop a real-time reconstruction algorithm for data acquired using such hardware. RGC operates at a single resonance frequency to create a sinusoidal gradient that can supplement the encoding provided by traditional gradient coils. The use of such high-frequency wave- like encoding (wave-RGC) during data readout have been shown to enable very high accelerations. Nonetheless, current wave-RGC efforts require calibrations through expensive field probes to capture spatiotemporal field non-idealities, and computationally intensive reconstruction. A key innovation in this proposal is to overcome these issues using i) a concept we termed ‘‘Implicit Representation of GRAPPA Kernels’, which we combined with ii) an observation that a family of GRAPPA kernels can be used to represent the spatiotemporal field of wave-RGC; an idea akin to ESPRIT for coil sensitivities. Ultlizing these ideas, a rapid GRAPPA-like calibration scan will be developed for use to train a Multilayer perceptron to implicitly represent a family of GRAPPA kernels, from which high-fidelity estimates of wave-RGC’s spatiotemporal field can be obtained. A k-space interpolating algorithm will then be developed to apply these GRAPPA-like kernels to transform highly-accelerated wave-RGC data to densely sampled cartesian data for real-time FFT-based SENSE reconstruction. To demonstrate the benefits of our rapid calibration and efficient reconstruction, they will be used to facilitate two-challenging new wave-RGC applications with high payoffs. First, wave-RGC will be deployed in SMS-EPI to enable 16x accelerated functional MRI and provide a dramatic step gain in the spatiotemporal resolution at which we can study brain function non-invasively. The second application is to create a flexible, form-fitting RGC suitable for rapid knee-imaging (and extremity). Such flexible design creates undesirable cross-subject field variations that our proposed rapid calibration can effectively calibrate, while allowing the RGC’s diameter to be small to dramatically reduces power requirement (scale as r5); enabling it to be driven using a low-cost amplifier.