A Diffusion-Prepared Reduced FOV Sequence for Prostate MRI Near Metallic Implants

Abstract

Purpose:

To enable diffusion weighted imaging in prostate patients with metallic total hip replacements in clinically feasible scan times for prostate cancer screening, and avoid distortion and dropout artifacts present in the conventionally used Echo Planar Imaging (EPI).

Methods:

A reduced FOV diffusion-prepared sequence that is robust to the B₀ inhomogeneities produced by total hip replacements was achieved using high RF bandwidth pulses and manipulation for stimulated echo pathways. The reduced FOV along the A/P direction was obtained using slice-select gradient reversal, and the prepared magnetization was imaged with a 3D RF-spoiled gradient echo readout. The sequence was validated in phantom experiments, in vivo in healthy volunteers with and without total hip replacements, and in vivo in patients undergoing a standard MRI prostate exam.

Results:

The proposed sequence is robust to shading and distortion artifacts that are encountered by standard diffusion-weighted EPI in the presence of moderate off-resonance. ADC estimates obtained by the proposed sequence were comparable to those obtained with diffusion-weighted EPI.

Conclusion:

Acquisition of distortionless diffusion weighted images of the prostate is feasible in patients with total hip replacements on conventional, whole-body 3T MRI, using a b-value of 800 s/mm² and nominal resolution of 1.7 × 1.7 × 4 mm³ in scan times of six minutes.

1 |. INTRODUCTION

Prostate cancer is the most common male cancer, with 1.4 million new cases reported globally in 2020¹. MRI has become a key element in the screening and diagnosis of prostate cancer, and diffusion-weighted imaging (DWI) is central to the specificity of these prostate MRI examinations^2,3. However, clinical DWI sequences use Echo Planar Imaging (EPI) readouts, which is sensitive to susceptibility artifacts due to magnetic field inhomogeneities from rectal gas or metallic hip implants⁴. Susceptibility causes significant image artifacts including geometric distortion, signal pileups, and signal loss, potentially leading to non-diagnostic diffusion images. With a 2.3% prevalence of total hip arthroplasty in adults above 50 in the US as of 2010⁵ and predicted to rise continuously⁶, this issue will become more common.

Off-resonance manifests as geometric distortion in EPI due to its low effective bandwidth in the phase encode direction. Multi-shot DW-EPI⁷ reduces, but does not eliminate the geometric distortion due to bowel gas. Multiple blip polarity EPI acquisitions can further reduce geometric distortion⁸, but is still insufficient near hip replacements due to the severity of the signal pileups from metal. EPI also exhibits image shading due to T2* decay, and low RF excitation bandwidth if a water-only excitation is used. In patients with a total hip replacement, it is only the EPI readout that is prone to these artifacts, with other image contrasts mostly unaffected. The prostate is sufficiently distant from the implant that conventional T2-weighted FSE and T1-weighted RF-spoiled gradient echo sequences exhibit little artifact. A comprehensive clinical solution to prostate DWI near hip replacements must address geometric distortion due to off-resonance, image shading due to T2* and low RF bandwidth combined with B₀ inhomogeneities, and have clinically feasible scan times.

Diffusion-weighted steady-state sequences⁹ can achieve high excitation bandwidths with distortionless readouts. However, these sequences have difficulty achieving the relatively high b-values that are desired for prostate imaging^2,3, and lose the steady state signal when bulk motion occurs, resulting in signal dropout.

Multispectral imaging (MSI) sequences are used for metal artifact correction¹⁰ with the goal of imaging tissues directly adjacent to the implant. DW-MSI sequences^11,12 are not required for this application since the prostate is distant enough from the implant that it is feasible for all the spins of interest to be excited in a single RF pulse (only one bin is required).

Prepared sequences that store diffusion-weighted magnetization on the longitudinal axis offer an alternative to achieve diffusion imaging without distortion from off-resonance. This is because the choice of image encoding method is flexible since it is independent from the diffusion preparation. For prostate imaging, Zhang et al. applied a diffusion preparation with a Cartesian, variable flip angle Turbo Spin Echo (TSE) readout¹³. M1-nulling was used to reduce effects of shot-to-shot motion induced phase but no phase navigator correction was applied.

Instead of using TSE to image the prepared magnetization, we employed a 3D RF-spoiled gradient echo (FLASH, SPGR, T1-FFE) readout. Since the prostate is not directly adjacent to the implant, spin echos are not necessary to counteract T2* dephasing, and a gradient echo readout with a sufficiently short echo time exhibits little T2* shading. A gradient echo readout has more flexibility in choosing a flip angle schedule when compared to pseudo-steady state FSE trains¹⁴, and permits more k-space samples to be dedicated to phase navigator sampling without compromising the main echo train. We also describe a novel, high RF bandwidth diffusion preparation that prepares an inner volume, and provides reduced FOV imaging when used in tandem with a stimulated echo readout. An initial version of this work was presented at the 2024 ISMRM annual conference¹⁵. A similar scheme for achieving reduced FOV was concurrently and independently developed in¹⁶.

A reduced FOV strategy is well-suited for imaging the small volume of the prostate, and reduces the encoding requirement for the phase navigator, as well as residual artifacts from subcutaneous fat and other high-intensity tissues that are outside the prostate region. The proposed sequence was demonstrated in vivo on conventional, whole-body MRI at 3 Tesla using a b-value of 800 s/mm² and nominal resolution of 1.7 × 1.7 × 4 mm³, with a scan time of 6 minutes. It was validated in phantoms, in vivo with healthy volunteers with and without metallic hip replacements, and patients undergoing a regular prostate exam.

2 |. METHODS

2.1 |. Reduced FOV Diffusion Preparation

The reduced FOV diffusion-prepared sequence with RF-spoiled gradient echo (FLASH) readout is shown in Figure 1. First, y-selective saturation pulses are used to attenuate magnetization outside the y FOV. However, a considerable amount of magnetization will continue to exist due to B₁ inhomogeneity. Furthermore, it will have a high signal amplitude relative to the diffusion-prepared magnetization. We alleviate this issue by using the stimulated echo readout to further suppress the outer volume. At the end of the diffusion preparation, diffusion-weighted magnetization is stored in the longitudinal axis. A diffusion-prepared strategy permits a distortionless readout and high RF bandwidth pulses. The twice-refocused preparation comprises an excitation, refocusing, and tipup pulses. The preparation reduces the FOV along y (the in-plane phase encode direction), and the readout is selective along z (the slice select direction).

Figure 1.

Sequence diagram showing the y spatial saturation pulses, reduced FOV diffusion preparation, and RF-spoiled gradient echo (FLASH) stimulated echo readout.

This strategy of using the preparation to reduce the y FOV avoids the use of 2D spatially selective pulses in the readout¹⁷, as these have longer durations compared to 1D pulses and result in increased echo times. A 3D readout is well-suited for prostate imaging since the imaging volume is small, and the encoding requirement for shot-to-shot phase correction is relatively little. A spin echo readout is not required for this application because the prostate is sufficiently far enough away from the implant that an RF-spoiled gradient echo readout with a short echo time exhibits little T2* shading.

Stimulated echo diffusion preparations apply a stabilizer gradient prior to the tipup pulse¹³, which reduces the SNR by a factor of 2×. A stimulated echo is necessary because diffusion gradients impart a random, spatially-dependent phase that changes each shot due to involuntary bulk motion and motion-sensitizing diffusion gradients. The stabilizer before the tipup ensures that the same net signal magnitude is stored in the longitudinal axis regardless of the random phase. The stabilizer must be balanced during the readout, which is a key component of the proposed reduced FOV scheme. Although the stabilizer ensures that the same net signal magnitude is stored, the random phase persists in the readout, meaning shot-to-shot phase correction is still required in the reconstruction.

The reduced y FOV is achieved by employing gradient reversal along the y direction in tandem with a stimulated echo readout. The yf-plot shown in Figure 2 illustrates the prepared volume, which is the overlap of the gray and blue bands from the excitation/tipup and refocusing pulses respectively. Signal from magnetization that is outside the prepared y FOV is suppressed by the stabilizer rephaser, which can be explained by considering the Extended Phase Graph (EPG) states¹⁸ for different regions in the yf-plot.

Figure 2.

Diagram showing the spins that are excited in yf space, drawn to scale with the RF pulses used in the final sequence. The prepared FOV is the crossing of the excitation (gray, positive slope), refocusing (blue), and tipup (gray, negative slope) bands. The tapering of the excitation bandwidth at the edge of the y FOV demonstrates the need for high RF bandwidth excitation and tipup pulses. Spatial saturation bands (red) from additional selective pulses before the preparation reduce residual fat outside of the y FOV.

Extended phase graphs represent the magnetization as transverse states (denoted F_k), and longitudinal states (Z_k). Suppose that the area of the stabilizer applies 2 cycles per voxel and that the additional y spatial saturation pulses in Figure 1 are neglected. At the end of the preparation, the diffusion-prepared magnetization that should be imaged lies in the Z₂ state, and magnetization that is outside both gray bands are in the Z₀ state. All low-order F states are assumed to be zero at the end of the preparation due to the spoiler. During the stimulated echo readout, the slice-select axis is z, so all magnetization in the yf-plot undergoes the excitation (assuming the pulse has high RF bandwidth). The Z₂ state (which contains the diffusion-weighted magnetization), is partially transferred into F₂⁻ by the excitation, and into F₀ by the stabilizer rephaser, where it forms a stimulated echo that can be encoded. The Z₀ state (which contains magnetization outside the prepared FOV), is transferred into F₀ by the excitation, and then into F₂⁺ by the stabilizer rephaser. This dephasing is sufficient to suppress non-fat species. Simultaneous slice time-multiplexing is a related method that manipulates EPG states to simultaneously rephase and dephase different regions of the volume¹⁹.

Due to the high proton density of fat, residual fat signal from outside the prepared FOV (shown in the yf-plot as a yellow line) may appear in the final image. If the refocusing pulses provide 180 ° rotations, fat outside the prepared y FOV will have a large residual longitudinal component in Z₀ at the end of the preparation. We reduced this artifact by employing multiple y spatial saturation pulses before the preparation, which in combination with the stimulated echo readout, sufficiently suppresses fat that is outside the prepared y FOV.

2.2 |. High RF Bandwidth Preparation

The main consideration for RF pulse design is the RF bandwidth of the preparation, which must be maximized to reduce B₀-dependent shading caused by the proximity to the metallic hip replacement. It is also desirable for the pulses to be short in duration to minimize the echo time of the diffusion preparation. The inherent symmetry of the preparation (excitation, refocusing, and tipup) provides considerable flexibility in RF pulse design.

Simulations of the off-resonance at 3T produced by a total hip replacement comprising a cobalt-chromium (CoCr, susceptibility 900 ppm) alloy head, and titanium alloy femoral stem (susceptibility 182 ppm)²⁰, are shown in Figure 3. The off-resonance maps were simulated by convolving the dipole field kernel with the susceptibility map of the implant^20,21,22. Assuming that the center of the femoral head is 10 cm from the center of the prostate, and that the centers are aligned in the S/I direction (in practice, the prostate would be slightly inferior), it is reasonable to expect ±500 Hz off-resonance variation after a linear shim correction. This total hip replacement material composition is a “worst case scenario” susceptibility²³, since stainless steel is rarely used²⁴. Presently in the United States, procedures are trending towards an increased fraction of ceramic implants²⁴. Metal-on-metal load bearing surfaces are currently unavailable in the United States due to their removal from the market by the FDA in 2014, but metal-on-polyethlyene load bearing surfaces using a CoCr femoral head continue to be available^24,25.

Figure 3.

Simulations of the off-resonance profile created by a hip replacement with a CoCr femoral head and titanium femoral stem at 3T. Red lines denote the center of the femoral head, and dashed black lines indicate a 6 cm extent where the prostate would be located. The spatially varying off-resonance, with and without an ideal linear shim correction are shown.

The normal prostate gland has a volume of approximately 3 cm × 3 cm × 5 cm = 30 cubic centimetres^26,27, so generally will fit within a 12 cm FOV in the A/P direction. The reduced FOV pulse should therefore be designed to prepare a 12 cm FOV along y with 1 kHz RF bandwidth.

We applied gradient reversal along the y direction to restrict the y FOV of the preparation. However, gradient reversal between the excitation and tipup creates a tapering of the bandwidth at the periphery of the y FOV. It is therefore desirable to maximize the RF bandwidth of the excitation and tipup to reduce the effect of the taper. The RF bandwidth of an SLR excitation pulse²⁸ is limited by the peak RF amplitude. The maximum B₁ of a linear phase SLR pulse can be reduced by approximately a factor of 2× using the root-flipping algorithm²⁹. Using a root-flipped pulse creates a non-linear phase slice profile, but the symmetry of the preparation (excitation and tipup) can be used to cancel this non-linear phase. Root-flipping also creates symmetrical pulses, which allows the phases of the slice profile to compensate even after the slice-select gradient is negated. The same root-flipped pulses were also used for spatial saturation outside the prepared FOV.

For refocusing, hard 180° pulses may be used, but these are sensitive to both B₁ and B₀. A hard 180° refocusing pulse with duration B_1max 15 μT, 0.8 ms duration has a 90% bandwidth of 300 Hz, well below the target RF bandwidth of 1 kHz. A twice refocused scheme can be used to benefit from the Carr-Purcell-Meiboom-Gill (CPMG) condition³⁰ and reduce the sensitivity to imperfect 180° refocusing pulses or B1 inhomogeneity. However, this dependence on CPMG when using hard pulses may cause non-uniform diffusion weighting that is sensitive to off-resonance due to mixing of stimulated and spin echos with different b-values. Instead of a hard refocusing pulse, we employed Wide, Uniform Rate, Smooth Truncation (WURST)³¹ refocusing pulses, which exhibit adiabatic behaviour and provided a 90% bandwidth of 1 kHz. Two WURST refocusing pulses were used in the preparation to compensate the non-linear phase imparted by the WURST pulse.

The CPMG condition is not satisfied with root-flipped excitation and WURST refocusing pulses. First, the excitation and tipup are selective along y, whereas the WURST pulses are non-selective. Second, the phase profiles of the excitation/refocusing pair do not have identical phase profiles that satisfy non-linear phase CPMG¹². However, the B₁ insensitivity provided by non-linear phase CPMG is not necessary since the WURST pulses achieve a refocusing flip angle close to 180°. The yf selection regions of the final pulses are drawn to scale in Figure 2.

2.3 |. RF Pulse Design

The WURST pulse has duration 6 ms, frequency sweep from −2 to 2 kHz, B_1max 16 μT, N = 40. The root-flipped SLR 90 excitation pulse had duration 2.1 ms, TBW 10, B_1max 15 μT, 90% bandwidth 4000 Hz, passband-ripple 0.01, stopband ripple 0.001²⁸. Root-flipping reduced the maximum B₁ 30 μT to 15 μT.

The pulses and their simulated magnetization profiles are shown in Figure 4. The saturation pulses were offset in the y direction ± 11 cm from the prescribed FOV, and each pair was repeated three times to reduce sensitivity to B₁ inhomogeneity.

Figure 4.

RF pulse profiles and magnetization profiles for root-flipped excitation and WURST refocusing pulses, compared to conventional choices of linear phase SLR excitation and hard refocusing pulses. Root-flipping reduces the peak B₁ by a factor of 2×, which enables a 4 kHz bandwidth excitation. For refocusing, a hard 180° refocusing pulse with peak B₁ 0.16 G and duration 0.8 ms has a 90% bandwidth of 300 Hz, and the flip angle is extremely sensitive to B₀. Employing a WURST refocusing pulse with duration 6 ms provides a 90% bandwidth to approximately 1 kHz.

2.4 |. Reconstruction

Coil sensitivities were estimated from the center 16 × 16 × 16 k-space region of the b = 0 s/mm² images using ESPIRIT³² after geometric coil compression³³ to 8 virtual coils. Reconstructions were performed on a workstation with 128 GB RAM, 24 core Intel Xeon Gold 5200R CPU, and a 24 GB NVIDIA Titan RTX GPU. The reconstruction time was approximately 3 minutes.

The reconstruction used a non-linear phase navigated least-squares reconstruction^34,35 with l1-wavelet denoising, which was solved with FISTA³⁶. Data were normalized so that the energy in the conjugate phase coil combination of the b = 0 s/mm² image was 1. The 3D wavelet regularization parameter was empirically tuned to 1 × 10⁻³.

2.5 |. Phantom Acquisition

Scans were performed on a 3T Signa Premier (GE Healthcare) system using a small anterior AIR coil (GE Healthcare, 21 channels), and spine coil embedded in the table.

A phantom consisting of a cylindrical water container surrounded by vials with varying concentrations of acetone and water to achieve a range of ADCs³⁷ was scanned. The phantom was scanned with and without a hip replacement (cobalt chromium alloy head, titanium stem) placed adjacent to the acetone phantom. The center of the femoral head was placed approximately 8 cm from the center of the water container, in the same orientation relative to the main magnetic field as would occur for an in vivo implant.

DW FLASH scan parameters: 30 cm FOV, matrix size 128 × 128 × 24, slice thickness 4 mm, TE prep 62 ms, b-values 0/800 s/mm², 1 NEX each, ETL FLASH 64, flip angle 10°, readout bandwidth ±65 kHz, FLASH TE 2 ms, FLASH TR 4 ms, TR 4000 ms, maximum gradient amplitude 70 mT / m.

In the phantom, multi-shot spin-echo EPI was acquired with the same FOV and matrix size, 4 shots, readout R/L, phase encode A/P, and TE 63 ms, to demonstrate geometric distortion and dropout artifacts that would appear with EPI near the hip replacement.

2.6 |. in vivo Acquisitions

To demonstrate the artifact from fat outside the reduced y FOV, b = 0 s/mm² images were acquired with and without y spatial saturation pulses in a healthy male volunteer (age 30) following IRB approval and informed consent. This volunteer was imaged with the root-flipped excitation and WURST refocusing pulses, with and without additional y spatial saturation pulses. Two different healthy male volunteers (aged 32, 34) were imaged with DW-FLASH and DW-EPI for image quality comparison. The shim volume was set to an 8 × 8 × 8 cm³ region centered on the prostate.

The scan parameters for diffusion-prepared FLASH were: 22 cm FOV, matrix size 128 × 128 × 24, voxel size 1.7 × 1.7 × 4 mm, TE prep 62 ms, b-values 0/800 s/mm², 1/3 NEX, ETL FLASH 16 + 60 = 76, elliptical ky-kz sampling pattern with corner cutting, phase navigator (first 16 echos) flip angle 4°, readout flip angle readout 12°, 40 shots per NEX, readout bandwidth ±65 kHz, FLASH TE 2 ms, FLASH TR 4.4 ms, TR 2200 ms, FLASH readout duration ~330 ms, diffusion direction L/R (gradient amplitude 70 mT/m), stabilizer area 1.5 cycles / voxel, scan time 6 minutes.

Flow compensated diffusion encoding was employed to reduce the amplitude of the imparted shot-to-shot random phase, and improve consistency in the standard patient population. The first 16 echos of the echo train were used as a phase navigator, which sampled the center 16 × 8 ky-kz lines with Ry = 4 and Rz = 2 each shot (encoding a 5 cm × 3.5 cm FOV in yz and reconstruction to 20 cm × 7 cm). This relatively high acceleration factor can be reconstructed with parallel imaging because of the reduced y FOV. The phase navigator resolution was 1.5 cm × 1.5 cm × 3 cm.

Three patient volunteers undergoing a regular prostate exam were scanned following IRB approval and informed consent. As part of the institutions’ standard clinical practice, patients underwent a micro-enema³⁸ prior to the exam to reduce bowel gas artifacts in DW-EPI. Two other healthy male volunteers (aged 58, 62) with total hip replacements (one unilateral ceramic, one unilateral metallic) were imaged with the same protocol. A vendor-provided multi-shot DW-EPI MUSE⁷ was acquired in all cases for comparison: 30 cm FOV, matrix size 192 × 192, voxel size 1.6 × 1.6 mm², slice thickness 4 mm, TE 65 ms, TR 3000 ms, 4 shots, b-values 50/800 s/mm²(1/4 NEX), all diffusion axes enabled simultaneously, frequency encode L/R, water-only excitation (RF bandwidth 50% FWHM 400 Hz), scan time 2:40. Four shots means that each phase encode blip traverses 4 ky lines, and 1 NEX means that each k-space location was sampled once. For the 4-shot, 4 NEX, each k-space location was sampled 4 times, for a total of sixteen 90° excitations.

A T2-weighted product 2D FSE scan was acquired for reference: matrix size 320 × 256, voxel size 0.6 × 0.7 mm², slice thickness 3 mm, ETL 16, TE 110 ms, TR 5000 ms.

3 |. RESULTS

3.1 |. Phantom Experiments

Figure 5 shows images of the acetone phantom acquired with multiple sequences, with and without the presence of a total hip replacement. The direction of the main magnetic field is perpendicular to the page. The full FOV 2D FSE exhibits little shading when the hip replacement (drawn in red) is introduced. Spin-echo EPI exhibits distortion and signal dropout in the central water container when the hip replacement is introduced. The low excitation bandwidth of spin-echo EPI also causes signal dropout in the agar phantom placed to the right of the acetone vials. The reduced FOV in the PE direction provided by gradient reversal and the stabilizer in the preparation are demonstrated. Since DW-FLASH employs a high RF bandwidth and distortionless readout, the introduction of the total hip replacement introduces some shading in the outer vials, but geometric distortion is absent and the signal level and ADCs in the large central water vial are unaffected. The mean ADC in the water vial acquired with DW-FLASH was 2.2 × 10⁻³ mm²/s, with and without metal. The ADC correspondence for all vials is shown in Supporting Information Figure S2.

Figure 5.

Demonstration of the distortionless reduced FOV preparation in an acetone phantom and some surrounding agar blocks, with and without the presence of a total hip replacement. The head of the hip replacement (drawn in red) is oriented with the main magnetic field in a configuration that mimics in vivo. A) Full FOV 2D FSE. B) Full FOV Spin-Echo EPI. The agar block to the right of the phantom has low signal due to the low excitation bandwidth and the shim volume being centered on the vials. C-E) Reduced FOV DW-FLASH b = 0, 800 s/mm² and ADC map.

3.2 |. Acquisitions in Healthy Volunteers

Figure 6 shows the effect of the stabilizer and selective saturation pulses on reduced FOV in b = 0 s/mm² images acquired in a healthy volunteer. Figure 6 A shows an image acquired with the y-selective T2 preparation without a stabilizer gradient. The inner region is T2-weighted and dark, and the outer region has high signal intensity due to being unaffected by the preparation. Introducing y-selective saturation pulses (Figure 6 B) reduces the signal intensity outside the prepared FOV, but there is still considerable residual signal due to B₁ inhomogeneity. Introducing the stabilizer in Figure 6 C suppresses the outer signal, but some residual signal from fat persists. Figure 6 D shows the result of applying of additional spatial saturation pulses on both sides of the y FOV, which reduces the residual fat outside the prepared y FOV.

Figure 6.

Images from a healthy volunteer demonstrating the reduced FOV provided by the stabilizer. Coils were combined with root-sum-of-squares without coil compression. A) T2-prepared image without stabilizer. EPG states of the magnetization at the end of the preparation in different y bands are shown. Since no stabilizer is applied, the magnetization resides in Z₀ at the end of the preparation. Only the inner region is T2-weighted. B) With saturation pulses outside the prepared FOV, the dynamic range is improved. A large amount of residual signal remains outside the FOV due to B1 inhomogeneity. C) Image acquired with a stabilizer. SNR is reduced due to the 2× signal loss from the stimulated echo. Magnetization outside the y band remains in Z₀, but within the y band magnetization finishes the preparation in Z₂. Due to the stabilizer rephaser, the Z₀ component is transferred to F₂ during the readout and is suppressed. Fat that lies outside the prepared FOV is not completely suppressed by the stabilizer rephaser due to its high proton density. D) The residual fat artifact is reduced using y selective saturation pulses.

Figure 7 shows diffusion-weighted images acquired with DW-EPI and DW-FLASH in two healthy volunteers. The low b-value images acquired with both DW-EPI and DW-FLASH have comparable image contrast. The apparent resolution of DW-FLASH is lower than the reference DW-EPI, but similar image features, such as the urethra (white arrow) of Volunteer 1, are visible in both EPI and DW-FLASH acquisitions.

Figure 7.

Comparison of DW-FLASH and DW-EPI images in two healthy volunteers. The b = 0 s/mm² image acquired with DW-FLASH has similar T2 and image structures in ADC maps when compared to DW-EPI. A) Volunteer 1. B) Volunteer 2.

3.3 |. Acquisitions in Patient Subjects

Figure 8 shows diffusion-weighted and high-resolution T2-weighted images from three patients undergoing a standard prostate exam. Lesions visible in DW-EPI are also present in DW-FLASH images. For all patients, fat has high signal intensity in the DW-FLASH images and a ADC less than 0.1 × 10⁻³ mm²/s, which indicates that the shot-to-shot phase correction method is effective.

Figure 8.

Images from patient volunteers without total hip replacements comparing DW multi-shot DW-EPI and DW-FLASH with reduced FOV. The corresponding T2-weighted image is provided for reference. A, B) Two patients with PI-RADS 5 lesions (A) white arrow, (B) yellow arrow. C) Patient with BPH nodule restricting diffusion (red arrow).

Patient 1 was found to have a lesion in the left apical peripheral zone consistent with PI-RADS 5, highly suspicious for clinically significant prostate cancer. The mean ADCs in the lesion for Patient 1 (white arrow) were 0.63 × 10⁻³ and 0.68 × 10⁻³ mm²/s with DW-EPI and DW-FLASH respectively. Patient 2 had known biopsy proven prostate cancer presenting after treatment with radiation and androgen deprivation therapy and was found to have a left anterior peripheral zone lesion consistent with residual prostate cancer. The mean ADCs in the lesion for Patient 2 (yellow arrow) were 0.61 and 0.62 × 10⁻³ mm²/s for DW-EPI and DW-FLASH respectively. Patient 3 had a lesion in the right transition zone consistent with a benign prostatic hyperplasia (BPH) nodule, PI-RADS 2, clinically significant cancer unlikely to be present. The mean ADC of the restricting diffusion areas in Patient 3 (red arrow) was 1.0 × 10⁻³ and 1.1 × 10⁻³ mm²/s for DW-EPI and DW-FLASH respectively. Multiple factors contribute to the apparent resolution of DW-FLASH being lower than the reference DW-EPI, which are addressed in the Discussion.

Figure 9 shows DW images acquired in two healthy volunteers with total hip replacements with different femoral head materials. In a volunteer with a left ceramic hip replacement (Figure 9 A), EPI exhibits large distortions outside the prostate region, and minor distortions inside the prostate itself since the shim volume is centered on the prostate. DW-FLASH achieves similar image contrast and ADC maps to DW-EPI. The main artifacts caused by the hip replacement is the failure of parallel imaging and shot-to-shot phase correction. The slice shown in Figure 9 A shows a moderate failure where phase correction errors caused a replica of the subcutaneous fat to ghost near the prostate. This artifact is dependent on the number of shots and the degree of involuntary motion in the volunteer. These artifacts are not apparent in the DW-FLASH image due to the reduced FOV suppressing outer signal, and improved robustness to involuntary motion from flow-compensated diffusion encoding.

Figure 9.

Images from volunteers with total hip replacements. A) Volunteer with a left ceramic total hip replacement. Off-resonance causes fat suppression failures, and poor phase correction creates residual ghosts from fat. B) Volunteer with a right metallic total hip replacement. Off-resonance from the implant causes dropout (white arrow) in DW-EPI due to the low excitation bandwidth, and geometric distortion (red arrow). DW-FLASH excites spins closer to the implant and obtains images without geometric distortion (beige arrow). The yellow arrow points to a probable BPH nodule which is visible in both the DW-EPI and DW-FLASH acquisitions.

In the second volunteer, a titanium total hip replacement creates off-resonance that causes signal loss in EPI in the vicinity of the total hip replacement (white arrow). In this scenario, the signal loss infringes slightly on the prostate and the residual off-resonance causes geometric distortion (red arrow). The DW-FLASH image has no geometric distortion and images the entirety of the prostate, including some adipose tissue surrounding the prostate that is closer to the implant. This demonstrates that the high bandwidth reduced FOV preparation is capable of exciting all the spins of interest. In the anterior transition zone a focus of restricted diffusion signal is present corresponding to a probable BPH nodule (yellow arrow).

4 |. DISCUSSION

The presented diffusion weighted imaging method is robust to image shading and off-resonance induced geometric distortions that are present in the widely used DW-EPI. It is capable of imaging spins in the vicinity of, but not adjacent to, a metallic total hip replacement. This sequence may also be applicable for cases with severe bowel gas artifacts when enema is not performed³⁸ or is unsuccessful, or for female pelvis imaging. We presented images acquired in volunteers with and without hip replacements, but only two hip replacements, which caused moderate artifact in DW-EPI were studied. Further assessment of the proposed sequence is required in a wider variety of implant compositions. Patients with bilateral total hip replacements would exhibit greater artifacts.

The proposed sequence enables a tradeoff between robustness to off-resonance and image quality in an ideal scenario. In patients where spins are on-resonant, the diffusion-prepared sequence will have worse SNR efficiency compared to DW-EPI. However, there is a small percentage of patients where DW-EPI fails, and the only available contrasts are T2-weighted and pre/post contrast-enhanced images. DWI is a critical part of a prostate cancer imaging protocol^2,3, and it is not suitable that some patients are missing diffusion-weighted images.

A relatively low b-value of 800 s/mm² was employed to ensure that the images would have sufficient SNR for presentation. While a six-minute scan time is relatively long, it is not atypical for prostate protocols to use 5–10 minutes acquiring DW-EPI^39,40,41. If it is known a priori that EPI will fail, then diffusion-prepared FLASH b = 0 / 800 s/mm² could be acquired in six minutes, and b = 1400 s/mm² in an additional six minutes.

DW-EPI can employ 2D spatial pulses¹⁷ to reduce the FOV in the phase encode direction and shorten the EPI readout echo spacing⁴³. While application of this pulse has been shown to reduce distortion artifacts in the prostate^44,45, 2D spatial pulses suffer from a low excitation bandwidth, and similar to the spectral-spatial pulse, do not excite all spins of interest in the presence of off-resonance. Since the 2D spatial excitation employs an EPI trajectory, by duality of excitation and reception¹⁷, the excitation profile exhibits warping in the slow, blipped, slice direction. In combination with slice-selective refocusing, the composite profile results in off-resonant spins being suppressed. This suppression of off-resonant spins is shown by simulation in Supporting Information Figure S3, S4 for a realistic pulse design with: blip-direction z, slice FOV 10 cm (permitting 20 slices), slice thickness 5 mm, y FOV 15 cm, 29 subpulses, bipolar trajectory, subpulse duration 0.55 ms, total pulse duration 16 ms, bandwidth 200 Hz / cm. Due to the bipolar design, this 2D pulse is prone to excitation ghosts at z FOV / 2. In the composite profile, spins that have 100 Hz off-resonance (creating a shift of 0.5 cm in the blip direction) will not be refocused, resulting in dropout.

An image acquired in a patient volunteer with a bilateral total hip replacement that created severe dropout and distortion artifacts in DW-EPI is shown in Supporting Information Figure S1. In this patient, DW-FLASH was acquired with a single 0.8 ms hard refocusing pulse in the diffusion preparation. This incurred some shading in the prostate close to the implant due to the low RF bandwidth of the hard pulse. The twice-refocused high RF bandwidth preparation was not acquired due to time constraints.

No fat suppression was employed because inversion recovery (STIR) is required to obtain uniform fat suppression in the magnitude of off-resonance produced by an implant¹⁰. STIR reduces the signal by ~30%, which we avoided because the scan time is already extended due to the SNR penalty of the simulated echo.

An FSE readout was not employed because a spin echo is not required to counteract T2* shading due to the prostate being sufficiently distant from the implant. All spins of interest could be prepared and imaged in a single excitation with adequately designed high RF bandwidth pulses while also providing a reduced FOV, which generally suffer from low excitation bandwidth. A benefit of a 3D RF-spoiled gradient echo compared to a 3D FSE readout is that the excitation pulses in the readout have greater RF bandwidths of 5–10 kHz. The high bandwidth of the 3D RF-spoiled readout is possible because the excitation flip angles are low. In contrast, FSE would employ pulses with 50% FWHM bandwidths of ~1 kHz. The moderate RF bandwidth of 3D FSE makes the slab select amplitude small and close to non-selective, which risks exciting signal outside of the volume of interest and introducing cusp artifacts⁴⁶.

2D-DWFSE PROPELLER is an alternative for distortionless diffusion-weighted imaging. DW-FSE PROPELLER was acquired in two volunteers, shown in Supporting Information Figure S5. The scan times and image quality of DW-FSE PROPELLER were comparable to the proposed diffusion-prepared method, requiring approximately five minutes for full coverage of the prostate. However, DW-FSE PROPELLER does not benefit from reduced FOV, and the vendor’s implementation has high SAR, making it likely for bulk motion to occur during the scan. No fat suppression was employed since an inversion pulse would be required to achieve robust fat suppression near the implants, reducing SNR and further extending the scan time.

Although fat is bright in the DWI image and has a low ADC which can confound lesion identification, the absence of fat suppression may be acceptable for prostate DWI because fat does not overlap with the prostate. The presence of fat can be used to verify that the shot-to-shot phase correction was successful, since the ADC of fat is 50 – 100× lower than that of free water⁴⁷.

The apparent resolution of the DW-FLASH image is lower than the DW-EPI comparison, for which there are two contributing factors. First, the stabilizer introduces motion sensitivity in the slice direction since it imparts 2 cycles/ 4 mm of dephasing. Sub-mm motion in the slice direction during the 350 ms echo train readout introduces phase offsets between inner and outer k-space and reduces the apparent resolution. Compared to abdominal imaging, respiratory motion causes little displacement in the prostate, approximately 3 mm in the A/P direction over the entire respiratory cycle⁴⁸. The short duration of the echo train reduces sensitivity to sub-mm motion. This effect limits the lowest practical slice thickness that can be achieved with this sequence, but respiratory triggering may alleviate this.

A second factor that affects apparent resolution is that a B₁ transmit inhomogeneity greater than 1.0 will reduce apparent resolution in the y and z directions since the signal modulation from the echo train will decay quicker than what was designed for the flip angle and ETL pair. For a given ETL, the flip angle was chosen to provide a signal at the final echo that is 10% the amplitude of the first echo. For prostate imaging, B₁ inhomogeneity is surmountable since the B₁ variation across the prostate is relatively little due to the small size of the prostate. With appropriate prescan and rapid B₁ mapping, this effect can be mitigated. The signal of the stimulated echo diffusion preparation is also sensitive to B₁, since the diffusion-prepared magnetization is measured through three excitation pulses, as opposed to the one excitation pulse used in DW-EPI.

5 |. CONCLUSION

Robust diffusion-weighted imaging in patients with certain metallic total hip replacements can be achieved in clinically feasible scan times of six minutes with a b-value of 800 s/mm² and nominal resolution of 1.7 × 1.7 × 4 mm³ by employing a reduced FOV, diffusion-prepared sequence.