A deep learning method for image-based subject-specific local SAR assessment.

PURPOSE: Local specific absorption rate (SAR) cannot be measured and is usually evaluated by offline numerical simulations using generic body models that of course will differ from the patient's anatomy. An additional safety margin is needed to include this intersubject variability. In this work, we present a deep learning-based method for image-based subject-specific local SAR assessment. We propose to train a convolutional neural network to learn a "surrogate SAR model" to map the relation between subject-specific B1+ maps and the corresponding local SAR. METHOD: Our database of 23 subject-specific models with an 8-transmit channel body array for prostate imaging at 7 T was used to build 5750 training samples. These synthetic complex B1+ maps and local SAR distributions were used to train a conditional generative adversarial network. Extra penalization for local SAR underestimation errors was included in the loss function. In silico and in vivo validation were performed. RESULTS: In silico cross-validation shows a good qualitative and quantitative match between predicted and ground-truth local SAR distributions. The peak local SAR estimation error distribution shows a mean overestimation error of 15% with 13% probability of underestimation. The higher accuracy of the proposed method allows the use of less conservative safety factors compared with standard procedures. In vivo validation shows that the method is applicable with realistic measurement data with impressively good qualitative and quantitative agreement to simulations. CONCLUSION: The proposed deep learning method allows online image-based subject-specific local SAR assessment. It greatly reduces the uncertainty in current state-of-the-art SAR assessment methods, reducing the time in the examination protocol by almost 25%.

Thermal noise variance of a receive radiofrequency coil as a respiratory motion sensor.

PURPOSE: Development of a passive respiratory motion sensor based on the noise variance of the receive coil array. METHODS: Respiratory motion alters the body resistance. The noise variance of an RF coil depends on the body resistance and, thus, is also modulated by respiration. For the noise variance monitoring, the noise samples were acquired without and with MR signal excitation on clinical 1.5/3 T MR scanners. The performance of the noise sensor was compared with the respiratory bellow and with the diaphragm displacement visible on MR images. Several breathing patterns were tested. RESULTS: The noise variance demonstrated a periodic, temporal modulation that was synchronized with the respiratory bellow signal. The modulation depth of the noise variance resulting from the respiration varied between the channels of the array and depended on the channel's location with respect to the body. The noise sensor combined with MR acquisition was able to detect the respiratory motion for every k-space read-out line. CONCLUSION: Within clinical MR systems, the respiratory motion can be detected by the noise in receive array. The noise sensor does not require careful positioning unlike the bellow, any additional hardware, and/or MR acquisition. Magn Reson Med 77:221-228, 2017. © 2016 Wiley Periodicals, Inc.

2D radially compensating excitation pulse in combination with an internal transceiver antenna for 3D MRI of the rectum at 7 T.

PURPOSE: The high precession frequency in ultrahigh field MRI coincides with reduced RF penetration, increased RF power deposition and consequently can lead to reduced scan efficiency. However, the shorter wavelength enables the use of efficient antennas rather than loop coils. In fact, ultrathin monopole antennas have been demonstrated at 7 T, which fit in natural cavities like the rectum in the human body. As the RF field generated by the antenna provides an extremely nonuniform B1 field, the use of conventional RF pulses will lead to severe image distortions and highly nonuniform contrast. However, using the two predominant dimensions (orthogonal to the antenna), 2D RF pulses can be designed that counteract the nonuniform B1 into uniform flip angles. In this study the authors investigate the use of an ultrathin antenna not only for reception, but also for transmission in 7 T MRI of the rectum. METHODS: The 2D radially compensating excitation (2D RACE) pulse was designed in matlab. SAR calculations between the 2D RACE pulse and an adiabatic RF pulse (BIR-4) have been obtained, to visualize the gain in decreasing the SAR when using the 2D RACE pulse instead of an adiabatic RF pulse. The authors used the 7 T whole body MR system in combination with an internally placed monopole antenna used for transceiving and obtained 3D gradient echo images with a conventional sinc pulse and with the 2D RACE pulse. For extra clarity, they also reconstructed an image where the receive field of the antenna was removed. RESULTS: Comparing the results of the SAR simulations of the 2D RACE pulse with a BIR-4 pulse shows that for low flip angles (θ < 41°) the SAR can be decreased with a factor of 4.8 or even more, when using the 2D RACE pulse. Relative to a conventional sinc excitation, the 2D RACE pulse achieves more uniform flip angle distributions than a BIR-4 pulse with a smaller SAR increase (16 × versus 64 ×). CONCLUSIONS: The authors have shown that the 2D RACE pulse provides more homogeneous flip angles for gradient echo sequences when compared to a conventional sinc pulse albeit at increased SAR. However, when compared to adiabatic RF pulses, as shown by simulations, the SAR of the 2D RACE pulse can be an order of magnitude less. Phantom and in vivo human rectum images are obtained to demonstrate that the 2D RACE pulse can provide a uniform excitation while transmitting with a single ultrathin endorectal antenna at 7 T. The combination of thin rectal antennas with efficient uniform transmit can open up new possibilities in high resolution imaging of rectal cancer.