Rapid estimation of 2D relative B1+‐maps from localizers in the human heart at 7T using deep learning

PurposeTo mitigate inhomogeneity at 7T for multi‐channel transmit arrays using unsupervised deep learning with convolutional neural networks (CNNs).MethodsDeep learning parallel transmit (pTx) pulse design has received attention, but such methods have relied on supervised training and did not use CNNs for multi‐channel maps. In this work, we introduce an alternative approach that facilitates the use of CNNs with multi‐channel maps while performing unsupervised training. The multi‐channel maps are concatenated along the spatial dimension to enable shift‐equivariant processing amenable to CNNs. Training is performed in an unsupervised manner using a physics‐driven loss function that minimizes the discrepancy of the Bloch simulation with the target magnetization, which eliminates the calculation of reference transmit RF weights. The training database comprises 3824 2D sagittal, multi‐channel maps of the healthy human brain from 143 subjects. data were acquired at 7T using an 8Tx/32Rx head coil. The proposed method is compared to the unregularized magnitude least‐squares (MLS) solution for the target magnetization in static pTx design.ResultsThe proposed method outperformed the unregularized MLS solution for RMS error and coefficient‐of‐variation and had comparable energy consumption. Additionally, the proposed method did not show local phase singularities leading to distinct holes in the resulting magnetization unlike the unregularized MLS solution.ConclusionProposed unsupervised deep learning with CNNs performs better than unregularized MLS in static pTx for speed and robustness.

“…We note that synthetic

{B}_1^{+}

maps have also been proposed in the literature, which would eliminate the curation of

{B}_1^{+}

map databases 23,35 . We also note newer efforts for circumventing the acquisition of

{B}_1^{+}

maps in the pTx pipeline by estimating it from

{B}_1^{-}

maps 36 . The synergistic combination of our approach with such techniques warrants further investigation.…”

Section: Discussionmentioning

confidence: 65%

“…23,35 We also note newer efforts for circumventing the acquisition of B + 1 maps in the pTx pipeline by estimating it from B − 1 maps. 36 The synergistic combination of our approach with such techniques warrants further investigation.…”

Section: F I G U R Ementioning

confidence: 98%

Unsupervised deep learning with convolutional neural networks for static parallel transmit design: A retrospective study

Kilic,

Liebig,

Demirel

et al. 2024

“…In contrast, the example hybrid approach 53 in Figure S1, albeit suffering from artifacts, took eight one‐channel‐on GRE scans of 6.1 s each and a single 12.6‐s all‐channels‐on DREAM acquisition (61.4 s total).

{\mathrm{B}}_1^{+}

mapping may be accelerated using improved implementations of such hybrid methods, 42 deep learning, 8,62,63 or undersampling 64 . There are also methods that avoid subject‐specific

{\mathrm{B}}_1^{+}

mapping altogether like plug‐and‐play MR fingerprinting 65 or the use of universal pulses, 66 as was done at 7 T in the heart 67 .…”

Section: Discussionmentioning

confidence: 99%

Whole‐liver flip‐angle shimming at 7 T using parallel‐transmit k_T‐point pulses and Fourier phase‐encoded DREAMB1+ mapping

Runderkamp,

Roos,

van der Zwaag

et al. 2023

PurposeTo obtain homogeneous signal throughout the human liver at 7 T. Flip angle (FA) shimming in 7T whole‐liver imaging was performed through parallel‐transmit kT‐point pulses based on subject‐specific multichannel absolute maps from Fourier phase‐encoded dual refocusing echo acquisition mode (PE‐DREAM).MethodsThe optimal number of Fourier phase‐encoding steps for PE‐DREAM mapping was determined for a 7T eight‐channel parallel‐transmission system. FA shimming experiments were performed in the liver of 7 healthy subjects with varying body mass index. In these subjects, first shimming and Fourier PE‐DREAM mapping were performed. Subsequently, three small‐flip‐angle 3D gradient‐echo scans were acquired, comparing a circularly polarized (CP) mode, a phase shim, and a kT‐point pulse. Resulting homogeneity was assessed and compared with estimated FA maps and distributions.ResultsFourier PE‐DREAM with 13 phase‐encoding steps resulted in a good tradeoff between accuracy and scan time. Lower coefficient of variation values (average [min‐max] across subjects) of the estimated FA in the volume of interest were observed using kT‐points (7.4 [6.6%–8.0%]), compared with phase shimming (18.8 [12.9%–23.4%], p < 0.001) and CP (43.2 [39.4%–47.1%], p < 0.001). kT‐points delivered whole‐liver images with the nominal FA and the highest degree of homogeneity. CP and phase shimming resulted in either inaccurate or imprecise FA distributions. Here, locations having suboptimal FA in the estimated FA maps corresponded to liver areas suffering from inconsistent signal intensity and T1‐weighting in the gradient‐echo scans.ConclusionHomogeneous whole‐liver 3D gradient‐echo acquisitions at 7 T can be obtained with eight‐channel kT‐point pulses calculated based on subject‐specific multichannel absolute Fourier PE‐DREAM maps.

“…

{S}_{\mathrm{image}}

can be utilized for a rapid evaluation of the imaging information contained in the OP mode 94 . In cases of strong deviations, alternative methods can be applied, such as dynamic pTx pulses, for example, kT‐points, based on rapidly acquired

{B}_1^{+}

‐maps using deep learning‐based approaches 95–99 . Linear combinations of other NMs could also be used to improve the imaging performance of a single NM.…”

Section: Discussionmentioning

confidence: 99%

“…94 In cases of strong deviations, alternative methods can be applied, such as dynamic pTx pulses, for example, kT-points, based on rapidly acquired B + 1 -maps using deep learning-based approaches. [95][96][97][98][99] Linear combinations of other NMs could also be used to improve the imaging performance of a single NM. This optimization could be performed based on acquired B + 1 -maps 40 or the sensor signal alone by using the S image of each NM for optimization.…”

Section: Ptx Imagingmentioning

confidence: 99%

Wirelessly interfacing sensor‐equipped implants and MR scanners for improved safety and imaging

Silemek

Seifert

Petzold

et al. 2023

PurposeTo investigate a novel reduced RF heating method for imaging in the presence of active implanted medical devices (AIMDs) which employs a sensor‐equipped implant that provides wireless feedback.MethodsThe implant, consisting of a generator case and a lead, measures RF‐induced ‐fields at the implant tip using a simple sensor in the generator case and transmits these values wirelessly to the MR scanner. Based on the sensor signal alone, parallel transmission (pTx) excitation vectors were calculated to suppress tip heating and maintain image quality. A sensor‐based imaging metric was introduced to assess the image quality. The methodology was studied at 7T in testbed experiments, and at a 3T scanner in an ASTM phantom containing AIMDs instrumented with six realistic deep brain stimulation (DBS) lead configurations adapted from patients.ResultsThe implant successfully measured RF‐induced ‐fields (Pearson correlation coefficient squared [R2] = 0.93) and temperature rises (R2 = 0.95) at the implant tip. The implant acquired the relevant data needed to calculate the pTx excitation vectors and transmitted them wirelessly to the MR scanner within a single shot RF sequence (<60 ms). Temperature rises for six realistic DBS lead configurations were reduced to 0.03–0.14 K for heating suppression modes compared to 0.52–3.33 K for the worst‐case heating, while imaging quality remained comparable (five of six lead imaging scores were ≥0.80/1.00) to conventional circular polarization (CP) images.ConclusionImplants with sensors that can communicate with an MR scanner can substantially improve safety for patients in a fast and automated manner, easing the current burden for MR personnel.