SAR-Efficient Sub-Volume Imaging Using Nonlinear Gradient Magnetic Fields Authors: Emre Kopanoglu, Ergin Atalar, R. Todd Constable Published: 2026 ABSTRACT Excitation using nonlinear gradient magnetic fields is investigated as a means of sub-volume magnetic resonance imaging (MRI). Conventional gradient fields provide encoding along a single direction, whereas nonlinear gradient fields encode information simultaneously along at least two directions. This leads to excitation regions (FOX) that have curvilinear boundaries, which may be more tolerant to aliasing artifacts when the encoded field of view (FOV) is smaller than the FOX. This reduces the complexity of the required radiofrequency (RF) excitation pulses and enables accelerated reduced-FOV imaging with standard slice-selection RF-pulses. We demonstrate the approach using a Z2-harmonic field for cylindrical regions of interest (ROIs) with various radius/height ratios. The minimum-FOV that should be encoded is formulated in terms of ROI and RF pulse parameters to allow a theoretical evaluation of feasibility during study design. The investigated method is compared to one-dimensional and two-dimensional selective RF pulses in terms of echo time, scan time and specific absorption rate (SAR) using simulations and phantom experiments. The investigated method yields lower scan time while keeping the SAR unaltered compared to a conventional slice-selective RF pulse, and is more efficient in terms of SAR, echo time and scan time compared to two-dimensional selective excitation. 1. Introduction MRI acquisitions are often long and necessitate balancing scan time against image resolution. For most applications, the region of interest is a sub-volume of the imaged subject. Imaging a smaller field of view can mean a lower number of phase-encoding steps, thereby leading to shorter scan times. Consequently, various FOV reduction techniques have been proposed. Phase encoding steps can be reduced by localizing the excitation to conform to the ROI. Standard slice selection uses a band-limited RF pulse applied in the presence of a linear gradient field (LGF) [1,2]. When the ROI is also smaller than the cross-section, excitation may be further localized via saturation, refocusing, or multi-dimensional selective pulses [1,2,3,4,5]. Saturation and refocusing approaches use additional RF pulses before and after the excitation, respectively, while multi-dimensional selective excitation utilizes a complex RF pulse to localize the excitation to non-rectilinear volumes. However, such additional/sophisticated pulses may increase the echo and/or repetition times, especially for rapid sequences such as fast low-angle shot (FLASH) [6], and increase RF power deposition, which may become prohibitive for SAR-limited sequences. While parallel-transmit arrays can mitigate SAR [7,8], feasibility remains limited at higher field strengths due to the frequency dependence of SAR [9]. Phase encoding steps can also be reduced via parallel imaging techniques [10,11,12] by exploiting spatially varying receive-coil sensitivity profiles, and these techniques can be combined with localized excitation to reduce scan times further. Although these techniques are highly effective in regions close to the coils, the efficacy drops in regions where coil sensitivities are comparable, such as around the center of the subject [11,13]. Another approach to reduce phase encoding steps is to use actively or passively generated magnetic field inhomogeneities with nonlinear spatial variations for localization. Active methods include adjusting the region of sensitivity using spatially inhomogeneous static magnetic fields [14,15] and using surface [16,17] or volume [18] gradient coils to disturb spin phase coherence in specific regions. Passive methods use ferromagnetic materials to distort the field [19,20]. However, nonlinear field variations generally lack the spatial uniformity delivered by their linear counterparts. Moreover, steering the FOX inside the subject to conform to different ROIs may require additional fields or physical movement of the coil or the subject. Furthermore, for the same dynamic range of field values, field variations generated by higher-order fields may be lower in amplitude in certain regions. In this case, obtaining sufficient dephasing may require longer gradient pulses, thereby affecting echo/repetition times for rapid sequences. NLGFs vary along at least two spatial directions [21], which has been used for locally improved resolution [22,23], diffusion MRI with higher gradient strength within the tissue of interest [24], improving reconstruction quality in accelerated acquisitions [25,26], excitation with reduced heating [27], and multi-dimensional excitation [28,29,30]. Because of the inherent capability of NLGFs for simultaneous multidirectional spatial encoding, a fourth approach for localizing excitation is to use nonlinear gradient fields (NLGFs) in tandem with RF pulses. A band-limited RF pulse can localize the excitation along at least two dimensions when used with an NLGF, while the localization in the presence of an LGF would only provide a single-dimensional localization. The achieved excitation profiles are curvilinear rather than rectilinear, which may be more tolerant to aliasing artifacts when the encoded FOV is smaller than the FOX and enable removal of these artifacts in post-processing. Ref. [31] used this method to image a cylindrical ROI; however, the reduction in total scan time was not explicit as it was compared to non-selective excitation of the ROI instead of one-dimensional (1D) or two-dimensional (2D) selective excitation. Furthermore, a loop coil was used to generate the NLGF, and therefore, the excitation region could not be steered inside the subject. While Refs. [21,32] used a second-order field that could be steered, the aliasing artifacts were eliminated using refocusing pulses, which affect echo time and SAR. Refs. [27,28,33,34] demonstrated multi-dimensional selective excitation with NLGFs. While replacing linear gradient fields with their nonlinear counterparts may reduce the SAR increase in multi-dimensional pulses [27,28], these approaches still lead to an overall increase in echo time and SAR, compared to the conventional slice-selective excitation. Here, we use a second-order (Z2) field for selective excitation with a simple sinc RF pulse [35,36]. Similar to [31], we image cylindrical regions using a reduced FOV while allowing controlled aliasing artifacts. In contrast to prior work, the second-order field permits steering of the excitation profile using linear gradient fields (LGFs), as demonstrated in phantom experiments. We further compare the proposed approach with one-dimensional and two-dimensional selective excitation in terms of SAR, TE, TR, and the number of encoding steps—comparisons not previously reported. Unlike earlier studies [21,27,28,32,33,34], no additional or sophisticated RF waveforms are required, and therefore, SAR remains identical to conventional slice selection. We also demonstrate, both theoretically and experimentally, that the radius-to-height ratio of the cylindrical ROI can be adjusted, a capability previously suggested but not demonstrated [32]. Finally, we formulate the minimum encoded FOV in terms of the ROI height and radius, providing an explicit measure of feasibility for other applications. Using simulations and phantom experiments, we show scan time reductions of up to 81.5% relative to one-dimensional selective excitation and SAR reductions of up to 91% relative to two-dimensional selective excitation. 2. Materials and Methods 2.1. Theory 2.1.1. Definitions In this section, we derive expressions for the encoded field of view (eFOV) obtained when a simple band-limited RF pulse is applied together with a trapezoidal gradient waveform that controls either a linear or a nonlinear gradient field. Because time-limited RF pulses cannot produce ideal excitation profiles with uniform excitation inside the ROI and zero excitation outside, we adopt the following pulse-design objectives:(i)Achieve excitation within the ROI with a predefined level of variability;(ii)Constrain excitation in the outer volume to a specified maximum level;(iii)Adjust the eFOV such that any aliased excitation entering the ROI remains below a specified threshold [37]. To minimize imaging time, aliasing is permitted within the eFOV but not within the ROI itself. (i)Achieve excitation within the ROI with a predefined level of variability; Achieve excitation within the ROI with a predefined level of variability; (ii)Constrain excitation in the outer volume to a specified maximum level; Constrain excitation in the outer volume to a specified maximum level; (iii)Adjust the eFOV such that any aliased excitation entering the ROI remains below a specified threshold [37]. To minimize imaging time, aliasing is permitted within the eFOV but not within the ROI itself. Adjust the eFOV such that any aliased excitation entering the ROI remains below a specified threshold [37]. To minimize imaging time, aliasing is permitted within the eFOV but not within the ROI itself. To simplify theoretical analysis in MRI, imaged subjects are commonly approximated as three-dimensional volumes with elliptical, circular, or rectangular cross-sections. In this work, we model the object as a circular cylinder with radius R, since a rectangular cross-section is an over-simplification whereas an elliptical cross-section introduces an extra variable without any significance on the discussion. The object is assumed to be aligned with the z-axis. The ROI inside this object is also modeled as a circular cylinder with radius r and length h, centered at a distance of r0 from the object’s center along the direction r0⃑ (Figure 1b,c). The distances from the center of the ROI to the top and the bottom of the object (or the maximum longitudinal distances that the RF coil is sensitive to) are denoted by L+ and L−, respectively. Other parameters that relate to the RF pulse and the target excitation profile are demonstrated in Figure 1, and summarized below: -δOV, the maximum excitation that may alias into the ROI (OV: outer volume); δOV, the maximum excitation that may alias into the ROI (OV: outer volume); -mROI, target excitation inside the ROI; mROI, target excitation inside the ROI; -δROI, maximum excitation variation allowed inside the ROI; δROI, maximum excitation variation allowed inside the ROI; -ΔfROI, frequency band inside which the excitation is between mROI±δROI, where the corner frequencies are given by fROI+, fROI−; ΔfROI, frequency band inside which the excitation is between mROI±δROI, where the corner frequencies are given by fROI+, fROI−; -ΔfOV, frequency band outside which the excitation is between ±δOV, where the corner frequencies are given by fOV+, fOV−; ΔfOV, frequency band outside which the excitation is between ±δOV, where the corner frequencies are given by fOV+, fOV−; -ζ=ΔfOV/ΔfROI, bandwidth ratio of the two bands for a symmetrical frequency spectrum. For an asymmetrical frequency spectrum, the bandwidth ratio can be defined separately for positive and negative frequency bands with respect to the center (carrier) frequency (fc): ζ+=(fOV+−fc)/(fROI+−fc) and ζ−=(fc−fOV−)/(fc−fROI−). These frequency bands are color-coded in Figure 1a and referenced in subsequent figures to facilitate visualization of excitation in the presence of nonlinear gradient fields. ζ=ΔfOV/ΔfROI, bandwidth ratio of the two bands for a symmetrical frequency spectrum. For an asymmetrical frequency spectrum, the bandwidth ratio can be defined separately for positive and negative frequency bands with respect to the center (carrier) frequency (fc): ζ+=(fOV+−fc)/(fROI+−fc) and ζ−=(fc−fOV−)/(fc−fROI−). These frequency bands are color-coded in Figure 1a and referenced in subsequent figures to facilitate visualization of excitation in the presence of nonlinear gradient fields. For clarity, we consider isotropic spatial resolution, noting that the formulations given in the next subsections readily extend to anisotropic resolution by expressing the eFOV values in terms of the number of pixels along each direction. 2.1.2. Field of View Formulation—Linear Gradient Fields Here, we assume L++L−>2R, reflecting typical human anatomy, and consequently, slice selection is performed along the length of the cylinder, z (Figure 1d). To obtain the target excitation level inside the ROI, ΔfROI should satisfy:(1)ΔfROI=γ2πGzh, where Gz is the z-gradient amplitude and γ/2π is the Gyromagnetic ratio. In this case, the encoded FOV should be:(2)eFOVz=ΔfOV+ΔfROI2=ζ+1h2, to allow aliasing into the eFOV, but not into the ROI (Figure 1e). In the transverse plane, aligning one of the encoding directions with r0⃑ (Figure 1b) maximizes allowable aliasing into the eFOV while ensuring that the aliased signal remains outside the ROI. Denoting these directions using x′ and y′, the minimum eFOV in the transverse directions can be derived as (derivations omitted for brevity):(3)eFOVx′=R+r2−r02,(4)eFOVy′=R+r+r0. Readout encoding incurs negligible time cost and is therefore typically performed along the longest spatial dimension to minimize total imaging time. We denote the longest eFOV dimension as maxeFOVx′, eFOVy′, eFOVz. For isotropic resolution, the minimum imaging time is proportional to(5)ttotal∝eFOVx′×eFOVy′×eFOVzmaxeFOVx′, eFOVy′, eFOVz. In Equation (5), the eFOV along the readout direction is canceled by the denominator, rendering the minimum imaging time proportional to the eFOV along the two phase-encoding directions. Note that the above formulations remain valid for ROIs shifted along z, since the encoded FOV can be shifted by changing the center frequency, fc. 2.1.3. Field of View Formulation—Nonlinear Gradient Fields Here, we use the Z2-field, which has a three-dimensional spatial distribution characterized by z2−(x2+y2)/2. The Z2-field can be steered inside the object to r0⃑r0=x⃑x0+y⃑y0+z⃑z0 using the LGFs and a center frequency adjustment:(6)Gz2(z−z02−(x−x0)22−(y−y0)22)=Gz2z2−x22−y22+Gxx+Gyy+Gzz+1γf0, where f0=γGz2−0.5 x02−0.5 y02+z02 is the Larmor frequency offset and Gx=Gz2x0, Gy=Gz2y0 and Gz=−2Gz2z0 are LGF amplitudes needed to steer the field. Consequently, the formulations that follow remain valid for regions of interest displaced from the isocenter. The Z2-field, a cross section of which is shown in Figure 1f, vanishes on two conical surfaces defined by z=±(x2+y2)/2. Spins located between either the zero-field surface and the z-axis i.e., z>x2+y2/2 experience a positive frequency shift, whereas spins located between the two surfaces experience a negative frequency shift. As a result, a simple band-limited RF pulse that restricts excitation along a single direction when applied with a linear gradient (Figure 1e) instead constrains excitation along both the longitudinal and radial directions when applied with the Z2 field (Figure 1g,h). To achieve the desired excitation within the ROI, the positive (fROI+) and negative (fROI−) cut-off frequencies should satisfy:(7)fROI− =−r22 γGz22π,(8)fROI+=h24 γGz22π. Due to the symmetry of the Z2 field distribution, the resulting excitation profile is cylindrically symmetric (Figure 1h). Although nonlinear gradient fields enable multidirectional localization using a simple band-limited RF pulse, the excitation region exhibits extensions along the zero-field conical surfaces. Because these surfaces are obliquely oriented relative to the transverse axes, the longitudinal extent of the excitation region changes as a function of transverse position (Figure 1g). On the other hand, using an nth-order field (n>1) leads to nth-order expansions/contractions in the excitation profile. For the Z2 field, the ratio of longitudinal distances along the z-axis corresponding to fOV+ and fROI+ scales as fOV+/fROI+, compared to the linear scaling of fOV+/fROI+ associated with LGFs. This effect is illustrated in Figure 1e,g, where the transition region between fROI+ and fOV+ is contracted for the nonlinear case, providing enhanced localization. To determine the minimum encoded FOV for the NLGF case, we consider the region z