Reducing PNS with Minimal Performance Penalties via Simple Pulse Sequence Modifications on a High-Performance Compact 3T Scanner

Abstract

One of the major concerns associated with high-performance gradients is peripheral nerve stimulation (PNS) of the subject during MRI exams. Since the installation, more than 680 volunteer subjects (patients and controls) have been scanned on a compact 3T MRI system with high-performance gradients, capable of 80 mT/m gradient amplitude and 700 T/m/s slew rate simultaneously. Despite PNS concerns associated with the high-performance gradients, due to the smaller physical dimensions of the gradient coils, minimal or no PNS sensation was reported with most pulse sequences. The exception was PNS reported by only five of 252 subjects (about 2%) scanned with a specific 3D fast spin echo pulse sequence (3DFLAIR). Rather than derating the entire system performance across all pulse sequences and all gradient lobes, we addressed reported PNS effect with a simple and specific modification to the targeted lobes of the problematic pulse sequence. In addition, the PNS convolutional model was adapted to predict sequence-specific PNS threshold level and its reduction after derating. The effectiveness of the targeted pulse sequence modification was demonstrated by successfully re-scanning four of the subjects who previously reported PNS sensations without further reported PNS. The pulse sequence modification did not result in noticeable degradation of image quality or substantial increase in scan time. The results demonstrated that PNS was rarely reported on the compact 3T, and when it was, utilizing a specific modification of the gradient waveform causing PNS was an effective strategy, rather than derating the performance of the entire gradient system.

1. Introduction

The performance of the gradient system is a core specification of an MRI scanner. Higher gradient performance can lead to improved image quality and reduced scan time. However, one of the major concerns associated with high-performance gradients is the physiological effect known as magnetostimulation, commonly referred to as peripheral nerve stimulation (PNS). Rapidly switching, strong gradient fields (i.e., time-varying magnetic fields) induces electric fields in the subject being imaged. Nerves can be stimulated if these time-varying electric fields have sufficient amplitude (Recoskie et al., 2010). To date, commercially-available, whole-body scanners are capable of maximum gradient performance of 80 mT/m gradient amplitude and 200 T/m/s slew rate. With recent advances in imaging gradient technology, even higher performance gradient systems (McNab et al., 2013; Van Essen et al., 2012) are now available on research-dedicated MRI scanners. However, PNS commonly limits the concurrent use of the full gradient strength and slew rate for in-vivo whole-body imaging.

Recent studies (Lee et al., 2016; In et al., 2018) demonstrated that PNS could be substantially reduced on a compact, head-only 3T MRI (Foo et al., 2018) even when the gradients are operated at 80 mT/m gradient amplitude and 700 T/m/s slew rate, simultaneously. This is mainly due to the reduced gradient coil size: inner diameter of 42 cm, compared to 58–73 cm inner diameter gradient coils typically installed in conventional, whole-body scanners. Since there is an inverse relationship between the gradient coil size and its PNS thresholds, full gradient performance can be available for in vivo MR imaging on the compact 3T system with relaxed PNS restriction. With the benefit of high-performance gradients, a 48% reduction in echo spacing in echo-planar imaging (EPI) sequence can be achieved on the compact 3T compared to a 60-cm bore, whole-body MRI operating at 50 mT/m and 200 T/m/s (Tan et al., 2016). This resulted in substantially reduced image distortion (Tan et al., 2016), reduced image burring in arterial spin labelling (Kang et al., 2020), improved efficiency in multi-shot EPI (In et al., 2018; In et al., 2020), and increased data samples per unit acquisition time for improved susceptibility weighted imaging (Shu et al., 2018).

A compact 3T scanner developed under an NIH-funded Bioengineering Research Partnership was installed at Mayo Clinic Rochester in February 2016 (Foo et al., 2018). From April 2016 through April 2020, more than 680 subjects (patients and healthy controls) have been successfully scanned under IRB-approved protocols with written informed consent. As part of the protocols, subjects were asked directly to report any discomfort or pain experienced during the scanning. PNS sensation was reported by five out of a total of 252 subjects scanned with a sagittal three-dimensional (3D) fast-spin-echo (FSE) CUBE fluid-attenuated inversion recovery acquisition (3D Cube-FLAIR, or generic name 3DFLAIR) with an identical imaging protocol. Derating the entire gradient system, e.g., modifying the configuration file, or changing from the First Level Control to Normal Operating mode for dB/dt can reduce PNS for sensitive subjects. However, those solutions are non-specific, and incur performance penalties across all pulse sequences used during the entire exam, greatly reducing the benefit of high-performance gradient systems.

As an alternative approach, a specific pulse sequence modification is suggested in this work, which derates specific gradient lobes. The purpose of this study is i) to reduce or eliminate PNS sensation with the modified sequence in even the most sensitive of subjects and ii) to minimize loss imaging performance across all the pulse sequences used during the exam. Problematic gradient lobes were identified with both the PNS convolution model (Schmitt et al., 1998) and the in-vivo experiments, and then selectively derated with a simple pulse sequence modification. The effectiveness of the modification was demonstrated by re-scanning four of the subjects who had previously reported PNS sensation on the compact 3T scanner.

2. Methods

The original 3D Cube-FLAIR (GE Healthcare, Chicago, IL, USA) sequence includes slice-encoding lobes on the logical z-axis (slab selection and encoding) combined with gradient crusher lobes. (Fig. 1). The combined lobes utilize the maximum amplitude gradient and slew rate. In our experience, this gradient waveform has not caused reported PNS on whole-body systems running at 50 mT/m maximum amplitude and 200 T/m/s maximum slew rate. However, 5 of 252 subjects reported PNS on the compact 3T running at 80 mT/m and 700 T/m/s when scanned in the sagittal plane. The imaging parameters for the 3D Cube-FLAIR were: TR/TE/TI = 7600/94/2063 ms, in-plane and through-plane acceleration factor = 2, readout receiver bandwidth = ±62.5 kHz, slice thickness = 1.2 mm, echo train length = 200, FOV = 240×240 mm², matrix size = 256×256. The scan time was 6 minutes 46 seconds. As previously reported, the physical x-gradient coil, i.e., the axis oriented along the patient’s right/left, is the most sensitive axis for producing PNS on a head gradient (Schmitt et al., 1998; Lee et al., 2016). We first confirmed this by acquiring the 3D Cube-FLAIR with and without the gradient lobes on the physical x-gradient coil.

Figure 1.

3D-FSE (Cube) FLAIR sequence diagram. The abbreviations in the sequence diagram are: RF: radio frequency, SS: slice selection, RO: readout, PE: phase-encoding, SR: gradient slew rate, AMP: gradient amplitude, ESP: echo spacing, T2prep: T2-weighted preparation RF pulse. The crusher gradient is combined with PE gradient on the SS axis.

Next, a PNS convolution model (Schulte and Noeske, 2015) was utilized to predict the specific gradient lobes that represent an area of concern for PNS, the PNS simulation level for the problematic gradient waveform, and its reduction after derating. This is given by the convolution of the slew rate S_i(t) = dG(t)/dt of the gradient trajectory G_i(t) with the nerve response

$$R_i=\frac{1}{S_{min}}{\int }_0^t\frac{S_i\left(\theta \right)c}{{\left(c+t-\theta \right)}^2}d\theta ,$$ [1]

where i = (x,y,z) indicates the gradient axis and c is the chronaxie time constant. The stimulation slew rate S_min is the minimum slew rate to induce PNS at any gradient amplitude (Chronik and Rutt, 2001) and is determined by experimentally. A chronaxie time constant of c = 334 μs (Schulte and Noeske, 2015) and the minimum slew rate of S_min = 161 T/m/s measured on the most sensitive physical x-gradient coil (Lee et al., 2016) were applied for this simulation. The sequence diagram was obtained from the compact 3T scanner with the imaging protocols identical to the in-vivo imaging. Compared to the previous study (Schulte and Noeske, 2015), where the overall PNS threshold in percentage was obtained after combining all the gradient axes, the PNS threshold P_i,thresh was calculated separately on each gradient axis, which is given by

$$P_{i,\mathit{\text{thresh}}}=100\%\times \left|R_i\right|.$$ [2]

After pinpointing the source of the PNS to be the combined crusher/encoding lobe ramps on the slice selection axis, a simple derating option was implemented in the sequence to allow a user-specified degree of slew rate derating specifically for the ramps of these lobes only. The total area under crusher gradient lobe was preserved, so the lobe duration was increased correspondingly. In addition, none of the other gradient lobes was derated, so that the maximal 80 mT/m amplitude and 700 T/m/s slew rate was still available for them.

Four (of the original five) volunteer subjects who reported PNS sensation in their previous exams were available to be rescanned with the modified derated sequence in accordance with the IRB-approved protocol. As with standard clinical scanning, all subjects were instructed before the exam to alert the technologist if they experienced any discomfort, and were provided with a squeeze-ball alarm. The PNS threshold and corresponding echo spacing were recorded for all the rescanned subjects as a function of the amount of slew rate derating. Once the threshold was determined for each individual, a PNS-free, sagittal 3D Cube-FLAIR was acquired using either a 32-channel (Nova Medical, Wilmington, MA) or 8-channel head coil (Invivo, Gainesville, FL, USA). Concomitant field compensation (Weavers et al., 2017; Tao et al., 2017) was applied during the scan for the asymmetric gradient design on the compact 3T.

A detailed illustration of the compact 3T system showing the dimensions relative to a patient, the locations of the magnet coils and gradient coil was reported in a previous study (Foo et al., 2018). For a quantitative comparison between the two image volumes without and with derating the gradient lobes, the image signal-to-noise ratios (SNR) was calculated in gray and white matter areas. An identical image mask was applied to calculate the mean signal value after co-registration between the image volumes and brain extraction using the FSL toolkit (http://fsl.fmrib.ox.ac.uk/fsl/). Note that the standard deviation of image noise outside the signal area was calculated before the image registration. Contrast to noise ratio (CNR) was also obtained by the SNR subtraction between the gray and the white matter areas. Finally, a board-certified neuroradiologist compared the 3D Cube-FLAIR images obtained with and without derating, i.e., without and with reported PNS.

3. Results

For the non-derated PNS test, sensation was reported for the sagittal acquisition of the 3D Cube-FLAIR, which corresponds to the crusher gradient waveforms being played on the physical x-axis (patient right/left). Two subjects reported PNS sensation in the nasal (or sinus) area and the other reported PNS sensation in the forehead area. To verify the suspected gradient lobes were the cause of the reported PNS, the scan was repeated with the gradient amplitude of the lobes was first set to zero, and no PNS was reported. The slew rate of the lobes was then decremented from its highest value in steps of 25%. The PNS threshold levels were observed at slew rates on the crusher lobe of 467, 350, 168, and 222 T/m/s for the four subjects, respectively. As the slew rate decreased, the corresponding echo spacing increased accordingly, e.g., by a small amount on the order of few hundred microseconds (Table 1). Notably, the imaging protocols and the scan time remained unchanged, even with the small echo space penalty.

Table 1.

Crusher gradient slew rate vs. echo spacing (esp). Note that the default echo spacing (i.e. when SR=700 T/m/s) varies with the subject’s weight and scanner software revision. First two (1 and 2) and the following two subjects were scanned on the scanner with software revision of DV25 and DV26, respectively.

Subject	Crusher gradient slew rate [m/T/s] (Reduction ratio)							Subject weight [kg]
	667 (1.0)	534 (1.25)	445 (1.5)	381 (1.75)	334 (2.0)	222 (3.0)	168(4.0)
Subject 1 esp [μs]	3472	3624	3680 (*)	-	-		-	73
Subject 2 esp [μs]	3616	3680	3736	3800	3856 (*)		-	81.65
Subject 3 esp [μs]	3552	3616	3672	3736	3792	4032	4320 (*)	72.57
Subject 4 esp [μs]	3440	3504	3560	3624	3680	3920 (*)		54.43

^(*) indicates the maximum threshold level without PNS sensation measured on each subject.

Figure 2 illustrates the PNS simulation results based on the PNS convolution model. The readout, the phase-encoding, and the slice-encoding gradient waveforms were played on the physical z-, y-, and x-axes, respectively, in the sagittal 3D Cube-FLAIR scan (Fig. 2A). In the simulation, the highest PNS values were 54.0 (Figs. 2B–D), 106.86 (Fig. 2B), and 109.23% (Fig. 2C), respectively. The highest value on the physical x-axis was reduced by up to 86.17% (i.e. about 21.11% reduction) after derating the slew rate by 50% (Fig. 2D). Therefore, the highest PNS threshold of the crusher gradients and its reduction on the physical x-axis are consistent with the simulation results. On the y-axis, however, although the highest PNS threshold value was similar to the one on the x-axis, no PNS was reported in in-vivo experiments regardless of whether the crusher gradients were either turned off or derated, i.e., even without changing the phase-encoding gradients (Fig. 2D and Table 1).

Figure 2.

PNS simulation using the convolution model in the 3D Cube-FLAIR sequence, where the horizontal axis denotes the sequence readout time. In (A), the sequence diagram during the entire readout time (i.e. 200 echo train length) obtained on the scanner is presented. The readout (blue), the phase-encoding (green), and the slice-encoding gradient waveforms (red) are played on the physical z-, y-, and x-axis. In the zoomed areas indicated by the dashed rectangular boxes in (A), the highest PNS threshold on the physical y- (B) and z-axes without (C) and with the derated crusher gradients (D) are shown. Note that the slew rate of the crusher gradients decreased by 50% is presented in (D).

Figures 3 illustrates that there was no noticeable image quality difference observed in all datasets with and without the proposed sequence modification. Even with the increased echo spacing of up to 768 ms (~ 22%), both the image contrast and quality remained similar to those from the previous scan with PNS sensation. In a blinded comparison, equivalent image contrast and lesion conspicuity were reported by a board-certified neuroradiologist.

Figure 3.

3D Cube-FLAIR images acquired with (upper row) and without PNS sensation (bottom row) from three volunteers. For comparisons of the image quality and contrast, reformatted axial (Ax.), coronal (Cor.), and sagittal slices (Sag.) are chosen from a 3D brain volume. The time intervals between the scan with and without PNS sensation were 20-minutes (left column), 3-months (middle column), and 7-months (right column).

Figure 4 shows the image SNR and CNR of 3D Cube-FLAIR acquired with and without reported PNS sensation. In the first and second subject scans, the SNR and the CNR were reduced very slightly with the proposed sequence modification in both the gray and the white matter areas. The reductions in both the SNR and CNR were higher in the third subject with a very high derating factor of 4. Nevertheless, the SNR and CNR values were still acceptably high even with derating and the differences were minimal when the standard deviations were considered (Fig. 4B).

Figure 4.

Image SNR and CNR of 3D Cube-FLAIR images acquired with (gray) and without PNS sensation (black) from four volunteers, which is shown as a graph (A) and table (B). The CNR shows the SNR subtraction between gray matter (GM) and white matter (WM) areas. Note that a 3D Cube-FLAIR scan only without PNS sensation was performed on the fourth volunteer due to the uncomfortable level of PNS sensation.

4. Discussion and conclusion

In this study, we performed an investigation of PNS on the high slew rate gradient system and reported that only 2% subjects scanned with 700 T/m/s reported PNS sensation with a 3D-Cube FLAIR scan. Instead of derating the entire gradient system, a simple sequence modification was developed to address and mitigate concerns related to the subject-dependent PNS. This targeted approach minimized the loss of the imaging performance on the compact 3T system, and we believe that this method is directly applicable to other high-performance gradient systems such as reported in (Foo et al., 2020). In conjunction with a PNS simulation, feasibility was evaluated by rescanning the PNS-sensitive subjects. The results show that the proposed approach could be a more effective strategy than simply derating the entire gradient system, i.e., all waveforms on all three axes.

Single-shot, 2D EPI is commonly considered the most common pulse sequence that causes PNS. In our experience, PNS resulting from EPI was reported less frequently than with sagittal 3D Cube-FLAIR. In EPI, 409 of 411 subjects reported no PNS. Only two of the five subjects who experienced PNS with 3D Cube-FLAIR, also experienced PNS from 2D EPI. Moreover, the PNS experienced from 2D EPI was readily removed either by setting the readout axis to be along the physical A/P direction, or by reducing the receiver bandwidth from ±250 to 32 kHz. Neither of these simple prescription modifications (swap phase frequency nor reducing readout bandwidth) was effective for reducing PNS for sagittal 3D Cube-FLAIR, since the gradient waveform on slice axis was the source of the problem. Therefore, this report has focused on 3D Cube-FLAIR rather than EPI, although the methods described here can be readily applied to EPI sequences, because reducing receiver bandwidth or swapping the phase frequency direction is not optimal. Other pulse sequences commonly used on the 680 subjects, without reported PNS or any special pulse sequence modification include 3D MPRAGE, 2D FSE, 2D/3D GRE, 2D FLAIR, and 3D SWAN.

Modification of the pulse sequence and protocol design to mitigate PNS sensation was performed based on the observed PNS dependence. Since the PNS threshold strongly depends on the gradient coil design (Zhang et al., 2003; Chronik and Rutt, 2001; Lee et al., 2016), the results from the PNS-sensitive subjects in this study were consistent with the previously reported observations (Lee et al., 2016) that the physical x-gradient coil (patient right/left) has the most sensitive PNS possibility, compared to y- or z-gradient coils. Using the spatial dependence and examining the pulse sequence waveform applied along the patient’s right-left axis for rapid gradient switching allow us to readily pinpoint the problematic gradient lobes.

The potential PNS threshold of the 3D-Cube FLAIR sequence was also simulated using the PNS convolution model (Schulte and Noeske, 2015). The simulation results were consistent with the in-vivo experiments on the physical x-axis (Figs 2C and 2D), but not on the y-axis (Fig. 2B). It may be because an identical S_min value in Eq. 1 was applied to estimate the highest PNS simulation value without considering the different level of the PNS threshold on each gradient coil. This demonstrated that the S_min values measured separately on each gradient coil should be accounted for a more realistic PNS simulation. Due to the measurement of S_min on the physical y-gradient coil was beyond the hardware limit (Lee et al., 2016) and a very low PNS simulation value was shown on the physical z-coil in the simulation (Fig. 2), it was sufficient to consider the PNS threshold only on the physical x-coil in the sequence modification for the compact 3T scanner. Although the PNS threshold did not directly correspond to the level of PNS sensation in every individual (Table 1), the model was still useful to predict the problematic gradients and its PNS threshold reduction after derating.

The measured PNS threshold levels varied for the four subjects (Table 1), demonstrating strong subject-dependence of the PNS effect. This result is consistent with previous studies (Ham et al., 1997; Chronik and Rutt, 2001; Lee et al., 2016; Saritas et al., 2013). All subjects were explicitly asked to report any discomfort or pain they experienced during the scanning. Therefore, we believe that there was negligible PNS feeling during the entire exam in the subjects who responded negatively. In addition, we were unable to find any common, outlying physical features among the PNS-sensitive subjects. We speculate that this may be because the nerve membrane dynamics are rather sensitive to even small anatomical details of the body (Davids et al., 2017), which may not be noticeable in the conventional anatomical imaging or body habitus. This could result in a large variation of the PNS threshold levels, even among the PNS-sensitive subjects. Given these unknowns, we think the empirical results presented here, do support that derating the entire gradient system is not a very efficient way to deal with the subject-dependent variations on PNS thresholds.

We believe that the PNS was observed on the compact 3T before the pulse sequence modification because it was using “stock” software, originally designed for a whole-body system. The original pulse sequence designers had no reason to restrict the slew rate performance of the crusher ramp when using a standard 200 T/m/s system, so the ramp time was minimized by default. This, however, caused PNS in 5 of among 252 subjects with a sagittal 3D Cube-FLAIR acquisition on the compact 3T running at slew rate of 700 T/m/s. As a practical approach, a specific modification of the major gradient component causing PNS was applied in this study to mitigate the problem. Since the PNS simulation model can show the PNS threshold of the imaging gradients and the reduction after derating, flexible imaging protocols with different level of PNS threshold (e.g., modest and extreme leading to the PNS threshold reductions of 20 and 40%, respectively) could be implemented in all the pulse sequences and be utilized for the subject who reports PNS sensation during the scan. This could address the PNS sensation that is highly subject-dependent and minimize the imaging capability loss over the entire exam on the high-performance system.

As compact 3T and related gradient technology progresses, we expect that more investigation into effective pulse sequence design strategies will help mitigate PNS concerns with little or no performance penalty. A recent study (Davids et al., 2018) demonstrated that the PNS issue can be resolved by increasing the PNS threshold. This approach requires an additional, high-frequency (10 kHz) pre-excitation gradient waveform be added just prior to the problematic gradient lobes (Davids et al., 2018). However, we believe that method may not be practical for the 3D-Cube FLAIR sequence investigated in this study. The PNS convolution model (Schulte and Noeske, 2015) can also be utilized to optimize the gradient shape balancing the slew rate and duration in order to further minimize both the possibility of introducing PNS and the performance penalty. However, that optimization should be performed by an exhaustive search for each trajectory. A more systematic evaluation is required to validate the effectiveness of this approach, which is beyond the scope of this study, but remains as an area for future study.

To date, more than 680 subjects have been successfully scanned on the high-performance compact 3T MRI system with 700 T/m/s slew rate. PNS was not reported with most sequences, and less than 2% out of a total of 252 subjects reported PNS sensation with a sagittal 3D-Cube FLAIR scan. However, since the level of PNS sensation was still tolerable in three of the four PNS-sensitive subjects, it was possible to perform a comparison study using a modified 3D-Cube FLAIR scan. With the proposed, targeted approach, we were able to successfully address the PNS concerns associated with the use of high-performance gradients with minimal performance or image quality impact in all the PNS-sensitive subjects who were re-scanned.