Besides long acquisition times and restricted spaces, acoustic noise

Transcription

1 ORIGINAL ARTICLE Parallel ImagingYBased Reduction of Acoustic Noise for Clinical Magnetic Resonance Imaging Eric Y. Pierre, MS,* David Grodzki, PhD,Þ Gunhild Aandal, MD,þ Bjoern Heismann, PhD,Þ Chaitra Badve, MD,þ Vikas Gulani, MD, PhD,*þ Jeffrey L. Sunshine, MD, PhD,þ Mark Schluchter, PhD, Kecheng Liu, PhD, and Mark A. Griswold, PhD*þ Objectives: The objective of this study was to demonstrate the feasibility of improving perceived acoustic comfort for a standard clinical magnetic resonance imaging protocol via gradient wave form optimization and validate parallel imaging as a means to achieve a further reduction of acoustic noise. Materials and Methods: The gradient wave forms of a standard T2 axial turbo spin-echo (TSE) sequence in head examinations were modified for acoustic performance while attempting to keep the total acquisition and inter-echo spacing the same. Parallel imaging was then used to double the inter-echo spacing and allow further wave form optimization. Along with comparative acoustic noise measurements, a statistical analysis of radiologist scoring was conducted on volumes from standard and modified sequences acquired from 10 patients after informed consent was obtained. Results: Compared with TSE, significant improvement of acoustic comfort was measured for modified-sequences quiet TSE and quiet TSE with generalized autocalibrating partially parallel acquisitions (P = andP = , respectively), and no statistically significant difference in diagnostic quality was observed without the use of parallel imaging. Conclusions: Standard clinical magnetic resonance imaging protocols can be made quieter through adequate gradient wave form optimization. In scans with high signal-to-noise ratio, parallel imaging can be used to further reduce acoustic noise. Key Words: MRI, acoustic noise, parallel imaging, pulse sequence optimization (Invest Radiol 2014;49: 620Y626) Besides long acquisition times and restricted spaces, acoustic noise is a major source of patient discomfort in clinical magnetic resonance imaging (MRI) sequences. 1 The sound pressure level (SPL) during an acquisition can be greater than 100 dba, 2Y4 necessitating the use of cumbersome ear-protective equipment. Despite the use of such devices, the acoustic noise is still a practical problem encountered often with patients in the clinic, especially with pediatric examinations and interventional settings. Acoustic noise has additionally been shown to interfere with functional MRI experiment results 5Y7 and can induce confounding brain uptake of fludeoxyglucose in simultaneous magnetic resonance (MR)-positron emission tomography experiments. 8 Received for publication December 11, 2013; and accepted for publication, after revision, February 27, From the *Department of Biomedical Engineering, Case Western Reserve University, Cleveland, OH; Siemens AG, Erlangen, Germany; Department of Radiology, University Hospitals of Cleveland, Case Western Reserve University, Cleveland, OH; Haraldsplass Deaconess Hospital, Bergen, Norway; Division of Biostatistics, Case Western Reserve University, Cleveland, OH; and Siemens Medical Solutions USA Inc, Malvern, PA. Conflicts of interest and source of funding: Supported by Siemens Healthcare, Erlangen, Germany. Authors Grodzki, Heismann, and Liu are employees of Siemens Healthcare and were not involved in the patient selection process or the data analysis. Reprints: Mark Griswold, PhD, Case Center for Imaging Research, University Hospitals, Bolwell Bldg B121, Euclid Ave, Cleveland, OH. mark.griswold@case.edu. Copyright * 2014 by Lippincott Williams & Wilkins ISSN: /14/4909Y0620 Veterinarian MRI studies have also shown that high-spl experiments can be harmful to companion or experiment animals. 9 One of the causes for the acoustic noise generated by an MR acquisition is fast gradient switching. 10 In the presence of a strong magnetic field, the current pulses in the gradient coils generate Lorentz forces, causing vibrations of the gradient coils themselves and also vibrations of other conductive structures such as the body radiofrequency coil through eddy currents. One approach to reduce the noise generated by the gradient coils involves enclosing the coils in a vacuum chamber, 11Y13 which has produced a reported final SPL in the scanner room of approximately 70 dba for some standard sequences such as gradientrecalled and spin echo at 1.5 T. 12 Another proposed approach is to mechanically rotate gradient fields, 14 coupled with radial coverage of k-space. New designs of gradient coil systems including windings performing active acoustic control have also been introduced. 15Y17 Alternatively, creating or modifying pulse sequences to avoid acoustic noise generation is an attractive opportunity that compliments hardware-based solutions while avoiding increased costs of manufacturing. One such solution is to time the ramping up and ramping down of the gradient wave forms so that the induced scanner vibrations cancel each other. 18 Another approach is to use lower gradient amplitudes and slew rates of the gradient wave forms. 19 By low-pass filtering the gradient, we can avoid vibration frequencies of the gradient coil, which produce high acoustic responses. However, fast clinical acquisitions with inter-echo spacing on the order of 20 milliseconds or less would see reduced benefits from the direct application of this approach. 19 Recently, the use of parallel imaging coupled with the redesign of gradient wave forms has demonstrated further reduction of acoustic noise in echo planar imaging. 20Y22 This reduction can be achieved by counterbalancing lengthened gradient wave forms with increased acquisition speed, thereby reducing acoustic noise without increasing acquisition time while increasing inter-echo spacing, although at the cost of signal-to-noise (SNR). The aims of this study are 2-fold: first, to extend similar principles of gradient wave form redesign and acquisition dead-time optimization to other fast, standard clinical MR sequences; second, to demonstrate the viability of these modifications to achieve acoustic noise reduction and improved patient comfort without significant cost to pathology visualization capability, imaging time, or sacrificing gradient efficiency. The underlying hypothesis of this study was that, under fixed constraints on acquisition time and image contrast, high-snr imaging systems can still trade off some of this SNR to acquire diagnostic quality images with significant reduction in acoustic noise. To validate this hypothesis, this study focuses on a standard T2- weighted axial turbo spin-echo (TSE) sequence in head examinations. Two complimentary approaches to dead-time and gradient wave form optimizations are investigated here: first, bandwidth (BW) and gradient wave forms are optimized for acoustic performance and minimal impact on image quality. Second, further gradient wave form optimization and acoustic noise reduction are tested with the additional use of parallel imaging. If successful, these modifications could allow adaptable acoustic noise reduction of fast clinical sequences depending on the available and/or desired SNR, without a change in acquisition time, image contrast, system hardware, or diagnostic outcome. 620

2 PI-Based Acoustic Noise Reduction for MRI FIGURE 1. Comparison of conventional sequence (dashed lines) with quiet sequence (solid lines). The reduction of analog-to-digital conversion time represents a BW increase. FIGURE 2. Comparison of conventional TSE sequence (top) with qtse-g with R = 2 (bottom). With R = 2, every other phase-encoding step (shaded in the standard sequence) is removed, yielding a doubled echo spacing. The total acquisition time and effective TE are kept similar between the

3 Pierre et al MATERIALS AND METHODS To reduce the acoustic noise of TSE sequences, the gradient wave forms were optimized with an automatic algorithm that extends any slope duration to its maximum and reduces the number of slopes to their minimum, resulting in a reduction of the acoustic frequencies generated by the gradient. In particular, spoiling and crusher gradient lobes were replaced by long rising or descending slopes while maintaining a similar crusher moment. Assuming a fixed total acquisition time, the reduction of the gradient slew rate is constrained by the fixed inter-echo spacing. Therefore, the first approach to further reduce the gradient slew rate was to reduce the readout sampling time, as illustrated in Figure 1. This results in a slight increase in BW and, therefore, a tradeoff between reduction of SPL and loss of SNR. The second approach, compatible with the first, was to make use of parallel acquisition techniques to reduce the echo train length by a reduction factor R. Keeping the acquisition time constant, the inter-echo spacing is increased by R, which allows further stretching of the gradient wave forms. This effectively represents an investment of parallel imaging acceleration in acoustic noise reduction rather than in imaging time reduction. A schematic diagram of such a modified sequence is presented in Figure 2. The proposed quiet sequence was tested by modifying a standard 2-dimensional T2-weighted TSE sequence. The readout BW was increased from 107 Hz per pixel in the standard protocol to 125 Hz per pixel in the quiet TSE (qtse) sequence prototype. The effective repetition time (TR) and echo time (TE) were modified from TR/TE of 5000/93 milliseconds to TR/TE of 5180/85 milliseconds. This resulted in only a 3-second increase in acquisition time, from 1 minute 37 seconds to 1 minute 40 seconds, and there was no change in the number of slices allowed for the acquisition. The parameters same as those of the qtse protocol were used with generalized autocalibrating partially parallel acquisitions (GRAPPA) 23 at R = 2 to yield the qtse-grappa (qtse-g) protocol. For both qtse and qtse-g protocols, the gradient slopes were maximally stretched as described previously. Signal-to-noise ratio maps were evaluated with the bootstrap method 24 for each sequence on the basis of water phantom signal and noise scans, using 100 random pixel permutations of the obtained noise image. The mean SNR value was computed for each map within the same large region of interest of the phantom. Using these protocols, in vivo studies were performed on a 3-T MAGNETOM Verio scanner (Siemens Healthcare, Erlangen, Germany) with a 12-channel head coil in patients undergoing a routine head MR examination. Informed consent was obtained from the volunteer before the start of the study in accordance with the local institutional review board protocol. A total of 10 patient examinations were performed (6 men, 4 women; age range, 21Y71 years), each comparing standard TSE images with qtse and qtse-g images. The image resolution ( matrix), number of slices (26), slice thickness (5 mm), and slice orientation were kept identical throughout the 3 acquisitions. The average and peak SPL values of the background noise, created by the cold-head pump and the ventilation among other sources, were measured with A-weighted frequencies using a 2238 Mediator sound level meter (Brueel & Kjaer GmbH, Bremen, Germany). The device was placed at the head position of the patient inside the bore, with the patient off the table, and measurements were averaged over 30 seconds, placed within the scanner room beyond the 5-G line. The SPL level for each sequence was then measured using the same apparatus. To assess acoustic comfort improvement, a total of 10 volunteers with no known hearing impediment (3 women and 7 men, aged 23 to 35 years) were each scanned with 5 different sequences: the TSE, qtse and qtse-g sequences, a very quiet ultrashort TE pointwise encoding time reduction with radial acquisition sequence, 25 as well as an echo planar imaging-diffusion tensor imaging sequence with b = 0 and b = All volunteers were given standard clinical ear protective equipment. Localizer parameters were kept identical across all scans. After each sequence, the volunteers were asked to give a comfort score from 1 to 10 (10 representing maximum comfort as if no sequence was playing and 1 representing extreme discomfort) immediately after the sequence finished playing. Each sequence was played twice, and the total series of 10 sequences was played in a randomized order for each volunteer. The compiled scores were analyzed using a randomized block analysis of variance (ANOVA). Multiple comparisons were adjusted with the Tukey method. Because some of the imaging parameters were modified, particularly readout BW, TR, and TE, it is important to assess whether images acquired with quiet protocols maintain image quality compared with images acquired with standard protocols. To this end, the following experiment was performed: a total of 7 image volume sets were assembled from each of the 10 patients. The first 2 sets compared qtse with TSE volumes in both left-right orders. Similarly, another 2 sets compared qtse-g with TSE volumes in both left-right orders. Finally, 3 sets were assembled with the same volume on the left and right, TSE:TSE, qtse:qtse, and qtse-g:qtse-g. All 70 volume pairs were randomized and presented in the same order to 3 trained radiologists blinded to the acquisition technique, with reader 1 being a neuroradiologist fellow, reader 2 being an experienced radiologist, and reader 3 being an experienced neuroradiologist. The readers were asked the following question: On a scale from j10 to 10, how much better is the image quality of the volume on the right compared with the volume on the left, with a positive score indicating superiority of the right volume, a negative score indicating superiority of the left volume, and 0 representing no difference in quality between left and right? The graphical user interface used for the reading allowed user navigation through the paired-volume slices and simultaneous image windowing of the 2 displayed images. To correct for the left-right bias, TSE:qTSE score was subtracted to the qtse:tse score and the result was divided by 2. The average across the readers of the resulting corrected scores was then computed for each patient. Corrected scores were calculated in the same way for the qtse-g:tse comparison. One-sample t tests were used to test whether the mean average reader scores differed from zero. The 95% confidence intervals (CIs) for the mean scores were also calculated. One-sample t tests and CI were also carried out using each reader s scores separately. A t test was used to test whether the average of the reader ratings across the patients differed from zero. One-sample t tests and CI were also carried out for each reader separately. To further assess any impact of the modified sequences on pathology visualization, a blinded randomized study was later conducted with an additional 4 patients (for a total of 14 patients), similarly to the study of Chandarana et al. 26 For each volume, 4 image quality parameters were rated on a scale of 1 to 5 (5 indicating the most desirable quality): overall image quality, image contrast, image sharpness, and any apparent image artifact, such as aliasing or motion artifact. The details of the scoring system are given in Table 1. The 42 volumes of TSE, qtse, and qtse-g images from the 14 patients were rated by the TABLE 1. Image Characteristic Overall image quality Image contrast Image sharpness Image artifacts Image Quality Parameter Scoring System Scoring System 1, unacceptable; 2, poor; 3, acceptable; 4, good; 5, excellent 1, unacceptable; 2, poor; 3, acceptable; 4, good; 5, excellent 1, unreadable; 2, extreme blur; 3, moderate blur; 4, mild blur; 5, no blur 1, unreadable; 2, extreme artifact; 3, moderate artifact; 4, mild artifact; 5, no artifact 622

4 PI-Based Acoustic Noise Reduction for MRI FIGURE 3. Signal-to-noise ratio maps from a water phantom using the bootstrap method. Under each map is displayed the mean SNR within the phantom. same 3 radiologists in the previous study, and the results were analyzed using randomized block ANOVA with Tukey adjustment for multiple comparisons. RESULTS The SNR maps are displayed in Figure 3. With respect to the standard TSE sequence, the qtse sequence showed a 6% reduction of mean SNR from 419 to 392, whereas the qtse-g sequence showed a 37% reduction of mean SNR from 419 to 266. This 37% reduction is consistent with the acquisition scheme: for R = 2, the SNR is divided by a factor of at least ¾2, which represents a reduction of 30%. An example of image comparison between the images acquired with standard TSE, qtse, and qtse-g is given in Figure 4. The respective average and peak A-weighted decibel measurements for standard TSE, qtse, and qtse-g protocols are displayed in Table 2. When the recording device was placed at the same position with respect to the bore with no patient on the table, repeated measurements showed no variation in average and peak SPL; therefore, no standard deviation is reported. The achieved reduction of average SPL for qtse and qtse-g were 5 dba and 13 dba, respectively. The corresponding acoustic comfort scores are displayed in Figure 5. The qtse sequence allowed a significant increase in average acoustic comfort score from 5.4 to 7 (P G ), whereas the qtse-g sequence showed FIGURE 4. Representative slices from 2 patients acquired with the reference TSE sequence (left column), the qtse sequence (center column), and the qtse-g sequence (right column). The measured average/peak SPL is shown as inset. 623

5 Pierre et al TABLE 2. Comparison of A-Weighted Decibel Values Between Sequences and Background Noise Sequence Type Average SPL (dba) Peak SPL (dba) Background qtse qtse-g Standard TSE a significant increase in average comfort score from 5.4 to 7.35 (P G ). The radiologist scorings for the paired-image comparisons are displayed in Table 3. Results of self-comparisons are summarized in the first column. Bias-corrected scores comparing qtse with TSE are displayed in the second column, where positive scores indicate better image quality for standard TSE. The mean score across all 3 readers was 0.61 (P = 0.13; 95% CI, j0.23 to 1.45). Bias-corrected scores comparing qtse-g with TSE are displayed in the third column, where positive scores indicate better image quality for standard TSE. The mean score was 2.41 (P G ; 95% CI, 1.83Y2.98). A graphical representation of the 95% CI of the average scores across all readers for each comparison types is displayed in Figure 6. The average image quality parameter scores for each reader and averaged over the 3 readers are compiled in Table 4. For readers 1 and 3, virtually no differences in scores could be found between the TSE and qtse images for all quality parameters. Reader 2 s scores for TSE image rated slightly higher than those for qtse images. This difference in average scores never exceeded 0.4 and was not statistically significant (adjusted P value of 0.21). Notably, the overall image quality was rated as slightly worse than good (3.9) for the TSE images and between good and average (3.5) for the qtse images. A box plot of scores averaged across the readers is given in Figure 7. On average across the readers and on the basis of a randomized block ANOVA analyses, significant differences between TSE and q-tse-g were found for the overall quality and artifact parameters (Tukey-adjusted P values of and , respectively). In addition, q-tse differed from q-tse-g with respect to overall quality and artifact (adjusted P values and , respectively). The contrast score for q-tse-g was also significantly different from both other FIGURE 5. Comfort score comparisons. A score of 10 indicates acoustic comfort as if no sequence was playing, whereas a score of 1 indicates extreme discomfort. 624 TABLE 3. Comparative Reader Study s Statistical Analysis All Techniques Compared With Themselves Reader 1 Reader 2 Reader 3 Average 0.35 T 0.40 (0.06 to 0.64) P = 0.02 j0.03 T 0.11 (j0.11 to 0.04) P = T0 V 0.11 T 0.14 (0.01 to 0.21) P = 0.04 qtse:tse qtse-g:tse j0.20 T 0.26 (j0.38 to j0.02) P = T 1.96 (j0.10 to 2.70) P = T 1.59 (j0.41 to 1.86) P = T 1.17 (j0.23 to 1.45) P = T 0.59 (j0.22 to 0.62) P = (3.33 to 4.57) P G T 1.25 (2.18 to 3.97) P G (1.83 to 2.98) P G Results show mean and standard deviation of the scores after self-bias correction, with 95% CI in parenthesis. Positive score show preference of the right volume over the left volume, on a scale from j10 to +10 (0 indicating equivalency). methods (adjusted P values = 0.05 for both analyses). No other differences were statistically significant. The median scores for the overall image quality and image contrast were good for the TSE images, slightly less than good for the qtse images, and between good and acceptable for the qtse-g images. The median scores for image blur and image artifact were slightly better than mild for the TSE images, mild for the qtse images, and slightly worse than mild for the qtse-g images. FIGURE 6. The 95% CIs for average scores across all readers for volumes compared with themselves (left), qtse volumes compared with standard TSE ones, and qtse-g volumes compared with standard TSE ones (right). Positive scores show preference for standard TSE in the last 2 cases.

6 DISCUSSION TABLE 4. Image Quality Parameter Scores TSE Reader 1 Reader 2 Reader 3 Average of 3 Readers qtse qtse-g T T T T T T 4.3 T 4.2 T 4.4 T T T T T T 3.1 T 3.3 T 3.2 T 0.7* 0.8* T 0.9 T T T 3.6 T 3.9 T 3.1 T T T T T T T 3.7 T 3.8 T 3.6 T *P G compared with TSE. P e 0.05 compared with TSE. P G compared with TSE. PI-Based Acoustic Noise Reduction for MRI The proposed optimization of gradient wave forms achieved significant improvement in acoustic comfort as demonstrated in Figure 5. Without the use of parallel imaging, the 10% increase in BW coupled with efficient gradient wave form modifications noticeably reduced SPL by 9.2 dba. The image preference study shows that the slight cost in SNR introduced by these modifications does not cause the image quality to differ statistically significantly (Table 3). This result is further confirmed by the rating of image quality parameters (Table 4; Fig. 7). One reader gave slightly lower quality parameter scores for the quiet sequence images than for the standard sequence images, which can be imputed to the 6% reduction in average SNR. However, none of these differences were statistically significant. Furthermore, the other 2 readers rated overall image quality, image contrast, image sharpness, and image artifacts as equivalent between the 2 sequences. The additional use of parallel imaging allowed a total reduction in average SPL of 19.8 dba (Table 2): the average SPL of the modified sequence and the background noise was only 19.7 dba above the background noise, compared with 39.5 dba with the standard TSE sequence. This SPL reduction came at the cost of a 37% reduction in average SNR (Fig. 4). This noticeable decrease in SNR that explains the preference of radiologists for standard TSE images over the modified sequence with parallel imaging was significant (Tables 3 and 4). Nonetheless, the contrast and overall image quality obtained with such modified sequences remained above acceptable for diagnostic purposes. In addition, the mild to moderate blur and artifacts of such images did not represent a severe degradation from the mild blur and artifacts of the original TSE images. The images were acquired using a 12-channel head coil, and the observed degradation in SNR and image FIGURE 7. Box plots of image analysis scores averaged across the readers for TSE, qtse, and qtse-g images. 625

7 Pierre et al artifacts from the use of parallel imaging could be mitigated by using a higher number of channels. It must be noted that the SNR of the standard TSE images was rather high (Fig. 4) and that the impact on image quality might become more significant with an acquisition parameter set or a system with lower SNR. However, these results prove that, in this configuration, gradient wave form redesign, especially in combination with parallel imaging, is a viable solution to reduce SPL of standard clinical sequences while maintaining acceptable image quality for diagnostic purposes. In conclusion, gradient modifications coupled with parallel imaging can achieve more than a reduction of a factor of 10 in acoustic noise with standard, fast MR sequences, allowing improved patient comfort without significant degradation of imaging time or any hardware modifications. Without the use of parallel imaging, the same modifications can achieve noticeable reduction of acoustic noise for the patient while maintaining diagnostic quality. REFERENCES 1. Quirk ME, Letendre AJ, Ciottone RA, et al. Anxiety in patients undergoing MR imaging. Radiology. 1989;170:463Y Shellock FG, Morisoli SM, Ziarati M. Measurement of acoustic noise during MR imaging: evaluation of six worst-case pulse sequences. Radiology. 1994;191:91Y McJury M, Blug A, Joerger C, et al. Acoustic noise levels during magnetic resonance imaging scanning at 1.5 T. Br J Radiol. 1994;67:413Y McJury M. Acoustic noise levels generated during high field MR imaging. Clin Radiol. 1995;50:331Y Bandettini PA, Jesmanowicz A, Van Kylen J, et al. Functional MRI of brain activation induced by scanner acoustic noise. Magn Reson Med. 1998;39: 410Y Amaro E, Williams SCR, Shergill SS, et al. Acoustic noise and functional magnetic resonance imaging: current strategies and future prospects. J Magn Reson Imaging. 2002;16:497Y Tomasi D, Caparelli E, Chang L, et al. fmri-acoustic noise alters brain activation during working memory tasks. Neuroimage. 2005;27:377Y Chonde DB, Abolmaali N, Arabasz G, et al. Effect of MRI acoustic noise on cerebral fludeoxyglucose uptake in simultaneous MR-PET imaging. Invest Radiol. 2013;48:302Y Lauer AM, El-Sharkawy A-MM, Kraitchman DL, et al. MRI acoustic noise can harm experimental and companion animals. J Magn Reson imaging. 2012;36: 743Y Hedeen R, Edelstein W. Characterization and prediction of gradient acoustic noise in MR imagers. Magn Reson Med. 2005;37:7Y Katsunuma A, Takamori H, Sakakura Y, et al. Quiet MRI with novel acoustic noise reduction. MAGMA. 2002;13:139Y Edelstein WA, Hedeen RA, Mallozzi RP, et al. Making MRI quieter. Magn Reson Imaging. 2002;20:155Y Edelstein WA, Kidane TK, Taracila V, et al. Active-passive gradient shielding for MRI acoustic noise reduction. Magn Reson Med. 2005;53:1013Y Cho ZH, Chung ST, Chung JY, et al. A new silent magnetic resonance imaging using a rotating DC gradient. Magn Reson Med. 1998;39:317Y Mansfield P, Haywood B. Principles of active acoustic control in gradient coil design. MAGMA. 2000;10:147Y Haywood B, Chapman B, Mansfield P. Model gradient coil employing active acoustic control for MRI. MAGMA. 2007;20:223Y Li M, Rudd B, Lim TC, et al. In situ active control of noise in a 4 T MRI scanner. J Magn Reson Imaging. 2011;34:662Y Shou X, Chen X, Derakhshan J, et al. The suppression of selected acoustic frequencies in MRI. Appl Acoust. 2010;71:191Y Hennel F, Girard F, Loenneker T. Silent MRI with soft gradient pulses. Magn Reson Med. 1999;42:6Y De Zwart J, Vangelderen P, Kellman P, et al. Reduction of gradient acoustic noise in MRI using SENSE-EPI. Neuroimage. 2002;16:1151Y Witzel T, Wald LL. Methods for functional brain imaging Zapp J, Schmitter S, Schad LR. Sinusoidal echo-planar imaging with parallel acquisition technique for reduced acoustic noise in auditory fmri. J Magn Reson Imaging. 2012;36:581Y Griswold MA, Jakob PM, Heidemann RM, et al. Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn Reson Med. 2002;47:1202Y Robson PM, Grant AK, Madhuranthakam AJ, et al. Comprehensive quantification of signal-to-noise ratio and g-factor for image-based and k-space-based parallel imaging reconstructions. Magn Reson Med. 2008;60:895Y Grodzki D, Heismann B. Quiet T1-weighted head scanning using PETRA. In: Proceedings of the 21st Annual Meeting of ISRMRM. vol 21. Salt Lake City, UT: 2013: Chandarana H, Feng L, Block TK, et al. Free-breathing contrast-enhanced multiphase MRI of the liver using a combination of compressed sensing, parallel imaging, and golden-angle radial sampling. Invest Radiol. 2013;48:10Y