A Comparative Evaluation of 7T MRI for Epilepsy with Deep-Learning-Based Image Reconstruction and Dynamic Parallel Transmission.
Authors
Affiliations (2)
Affiliations (2)
- From the Department of Radiology (E.H.M., J.O.E., X.Z., S.T., V.N.P., E.M.W., J.V.M., V.G.), Department of Neurologic Surgery (E.H.M.), Mayo Clinic; Swiss Innovation Hub (E.H.M., T.Y., G.F.P.), Siemens Healthineers International AG; Department of Radiology (T.Y.), Lausanne University Hospital and University of Lausanne; LTS5 (T.Y.), Ecole Polytechnique Federale de Lausanne; MR Application Predevelopment (D.N., P.L.), Siemens Healthineers AG; Siemens Healthcare (J.H.), Erlangen. [email protected].
- From the Department of Radiology (E.H.M., J.O.E., X.Z., S.T., V.N.P., E.M.W., J.V.M., V.G.), Department of Neurologic Surgery (E.H.M.), Mayo Clinic; Swiss Innovation Hub (E.H.M., T.Y., G.F.P.), Siemens Healthineers International AG; Department of Radiology (T.Y.), Lausanne University Hospital and University of Lausanne; LTS5 (T.Y.), Ecole Polytechnique Federale de Lausanne; MR Application Predevelopment (D.N., P.L.), Siemens Healthineers AG; Siemens Healthcare (J.H.), Erlangen.
Abstract
7T MRI enhances lesion detection in epilepsy but is limited by radiofrequency transmission field (B1+) inhomogeneity and long scan times. Recent advancements in dynamic parallel transmission and deep-learning-based reconstructions offer promising solutions. We aimed to optimize an enhanced 7T epilepsy protocol incorporating these innovations and evaluate real-world benefits compared to standard 7T epilepsy protocol. We retrospectively compared 40 consecutive brain MRIs acquired using a standard 7T epilepsy protocol to 40 MRIs obtained with an enhanced protocol with dynamic parallel transmission and deep-learning-based <i>k</i>-space reconstructions. Quantitative metrics for comparison included image noise, signal homogeneity (coefficient of variation), and resolution/time trade-offs. The enhanced protocol demonstrated significant improvements in resolution, scan time, noise levels, and image homogeneity. Edge-enhancing gradient echo and magnetization-prepared rapid gradient echo with 2 inversions sequence exhibited a 57.8% reduction in voxel volume while reducing scan time by 33.0% and improving image homogeneity (<i>P</i>=.002) without a significant change in noise (<i>P</i>=0.09). Deep-learning-based reconstruction of coronal T2 turbo spin echo resulted in a 25.7% reduction in noise (<i>P</i><.001), and patient-specific B1+ shimming achieved homogeneity comparable to dielectric pads. Sampling perfection with application-optimized contrasts using a different flip angle evolutions fluid-attenuated inversion recovery had reduced noise (<i>P</i><.001), enhanced homogeneity (<i>P</i><.001), and halved voxel size while maintaining similar scan time. Deep-learning-based echo planar imaging susceptibility-weighted imaging improved acquisition time by 56.5% with a 20.5% reductionin noise (<i>P</i>=.001). Despite increased resolution and parallel transmission use, overall scan time was less than 25 minutes, half the duration recommended by the 7T Epilepsy Task Force. Integration of dynamic parallel transmission and deep-learning-based reconstructions enhances image resolution, reduces scan time, and improves image homogeneity, addressing barriers to routine clinical implementation of 7T MRI. These advancements may improve lesion conspicuity and contribute to better outcomes for patients with epilepsy.