A 3D dual-echo spiral sequence for simultaneous DSC- and DCE-MRI with single bolus injection

Abstract

Purpose:

Perfusion MRI reveals important tumor physiological and pathophysiologic information, making it a critical component in managing brain tumor patients. This study aimed to develop a dual-echo 3D spiral technique with a single-bolus scheme to simultaneously acquire both DSC and DCE data and overcome the limitations of current EPI-based techniques.

Methods:

A 3D spiral-based technique with dual-echo acquisition was implemented and optimized on a 3T MRI scanner with a spiral stair-case trajectory and through-plane SENSE acceleration for improved speed and image quality, in-plane variable-density undersampling combined with a sliding-window acquisition and reconstruction approach for increased speed, and an advanced iterative deblurring algorithm. Four volunteers were scanned and compared to the standard of care (SOC) single-echo EPI and a dual-echo EPI technique. Two patients were scanned with the spiral technique during a preload bolus and compared to the SOC single-echo EPI collected during the second bolus injection.

Results:

Volunteer data demonstrated the spiral technique achieved high image quality, reduced geometric artifacts, and high temporal SNR compared to both single-echo and dual-echo EPI. Patient perfusion data showed that the spiral acquisition achieved accurate DSC quantification comparable to SOC single-echo dual-dose EPI, with the additional DCE information.

Conclusion:

A 3D dual-echo spiral technique was developed to simultaneously acquire both DSC and DCE data in a single-bolus injection with reduced contrast usage. Preliminary volunteer and patient data demonstrated increased temporal SNR, reduced geometric artifacts, and accurate perfusion quantification, suggesting a competitive alternative to SOC EPI techniques for brain perfusion MRI.

INTRODUCTION

A critical aspect of brain tumor patient management is the radiographic assessment of tumor status, which is used for diagnosis, localization, surgical planning, and surveillance. Perfusion MRI measures the delivery of blood to the tissues, revealing important tumor physiological and pathophysiologic information, and therefore, is a critical component in managing brain tumor patients. There are two primary MRI techniques to measure perfusion using exogenous contrast agents: DSC- and DCE-MRI.¹ DSC-MRI is one of the most widely used physiologic imaging techniques in neuro-oncology, with a reported utilization rate of 85% in routine brain tumor scans at sites across the US and Europe.^2,3 DSC-MRI relates dynamic T₂* shortening to contrast agent pharmacokinetics and measures relative cerebral blood volume (rCBV).^4,5 DSC-MRI can differentiate glioma grades and tumor types, identify tumor components in non-enhancing glioma,^6,7 distinguish tumor recurrence from post-treatment effects,^8,9 and predict tumoral response and patient survival after targeted therapy^10,11. To a substantial but lesser degree, DCE-MRI is used in 40% of brain tumor scans, primarily in cases where DSC-MRI’s utility is confounded by susceptibility artifacts. DCE-MRI detects T₁ shortening caused by contrast agent pharmacokinetics, and measures the transfer constant, K^trans, a marker of vascular permeability. DCE-MRI can be used for differential diagnosis of brain tumors,^12,13 distinguish tumor recurrent from post-treatment radiation effects,¹⁴ and is prognostic of therapy response.^15,16

DSC- and DCE-MRI have distinct clinical advantages and challenges. An advantage of DSC-MRI is its ability to extract perfusion parameters that enable the characterization of infiltrative tumor in regions where the blood brain barrier is intact.^6,7 In terms of pulse sequences used for acquisition, DSC-MRI relies upon EPI sequences that are prone to susceptibility artifacts, confounding its use for lesions in the brainstem, cerebellum, inferior temporal and frontal lobes. Additionally, DSC-MRI is routinely used to evaluate recurrent lesions, which most commonly occur adjacent to surgical resection beds. In these cases, magnetic field distortions related to cranial fixation hardware and/or ventricular drains lead to similar artifacts. EPI-based sequences also frequently exhibit geometric distortions that hinder the accurate registration of DSC-MRI with the high-resolution 3D anatomic scans used for surgery, creating further challenges for the validated use of perfusion MRI to target tumor cell-rich brain regions.^17,18 In cases where EPI-related artifacts reduce the clinical utility of DSC-MRI, the conventional T₁-weighted imaging sequences used for DCE-MRI become a preferable alternative approach, as they are far less sensitive to susceptibility effects and provide improved geometric fidelity. DCE-MRI is primarily used to evaluate enhancing tumor volumes where the blood-brain barrier (BBB) is disrupted. However, this can limit its utility for peritumoral assessment and with drugs that may induce pseudoresponse (e.g., bevacizumab).

DSC- and DCE-MRI both provide complementary information on tumor hemodynamic and vascular status via administration of an exogenous contrast agent. A common approach to utilize the benefits of both DSC- and DCE-MRI is to collect both datasets during sequential injections of contrast agent. In this strategy, the DCE data are collected during the first bolus injection, and the DSC data are collected during the second bolus injection. Consensus recommendations for the DSC-MRI protocol for use in high-grade gliomas include the use of a full-dose preload, followed by a second full-dose bolus during DSC data collection with an intermediate flip angle (FA) and field strength-dependent TE.¹⁹ The dual-dose scheme provides the best overall accuracy and precision for estimating cerebral blood volume.¹⁹ However, this approach requires the administration of two doses, which raises contrast agent-related safety concerns, practical considerations, higher cost, prolonged scan times, and increased protocol variability²⁰. To address these concerns, a single-dose scheme for DSC with a low FA and field strength-dependent TE has been recommended, albeit with slightly decreased accuracy and precision.¹⁹ However, with the single-bolus scheme for DSC, clinical sites will be forced to choose between collecting either DSC or DCE data, ultimately limiting their clinical utility.

To reduce the contrast agent usage and address related safety and practical concerns while preserving the capability of collecting both DSC and DCE data, a single-bolus, 2D dual-echo approach has been proposed. This approach has demonstrated several advantages, including elimination of T₁ leakage effects, higher rCBV accuracy across a range of pulse sequence parameters, and more importantly, simultaneous acquisition of DSC and DCE data.²¹ The accuracy of 2D dual-echo DSC/DCE-MRI for robust rCBV mapping has been validated using a digital reference object²² and in patient studies²³. Nonetheless, one limitation of current dual-echo approaches is their reliance on EPI readouts, which are prone to susceptibility-induced distortion artifacts that may overlap with the location of brain tumors. Additionally, the use of EPI readouts necessitates long TRs (> 500 ms), which leads to insufficient T₁ weighting and consequently reduces the sensitivity of this approach for robust DCE-MRI use.

In MRI, off-resonance effects arise from main magnetic field inhomogeneity, susceptibility variation at tissue interfaces, surgical cavities, and metal implants. In spiral MRI,^24,25 these off-resonance effects manifest as blurring rather than warping, as in EPI. As a result, geometric accuracy is preserved, enabling more accurate co-registration with high-resolution anatomic scans used for surgery and radiotherapy. Additionally, dynamic B₀ field drift due to breathing or coil heating creates imaging blurring rather than spatial shift and has a smaller impact on the time series data. Spiral acquisition also allows for shorter TE (< 1 ms) compared to EPI (approximately ~8 ms). This helps improve the estimation of contrast agent-induced T₁ changes and consequently yields more reliable DCE data and accurate characterization of the arterial input function (AIF). Two-dimensional spiral sequences with a dual-echo or combined spin- and gradient-echo acquisition have been proposed and demonstrated advantages compared to EPI scans.^26–28 However, these 2D single-shot spiral scans typically require a very long spiral readout, posing challenges in image deblurring, especially at higher magnetic fields and/or higher resolutions. Additionally, the longer TR required for a larger spatial coverage reduces image T₁ weighting and temporal resolution. Recently, advancements in pulse sequence and trajectory design, along with deblurring algorithms, have improved the image quality of spiral MRI.^29–31 In this study, we propose a 3D dual-echo spiral method to explore these improvements for accurate DSC/DCE-MRI with a single bolus. A 3D acquisition provides high SNR and contiguous slices with an improved profile. In addition, a multi-shot approach decouples the choice of TEs (to optimize R₂* sensitivity) from the choice of spatial resolution and coverage and exhibits fewer coherent flow and motion artifacts compared to multi-shot EPI. Furthermore, the short TR and the very short TE of the first spiral echo also allow for more accurate characterization of T₁ changes needed for DCE-MRI. This benefit can be partially demonstrated through the maximum measurable contrast agent concentration³², which is higher for the 3D spiral sequence with a short TR than a 2D sequence with a long TR (as illustrated in Supplemental Fig. S1). In this work, the 3D dual-echo spiral sequence was implemented and tested on healthy volunteers to evaluate image quality and on two tumor patients to assess its accuracy.

METHODS

Pulse Sequence and Image Reconstruction

A multi-shot 3D dual-echo spiral technique with a spiral-out readout gradient waveform (Fig. 1a) was developed on a Philips Ingenia 3 T scanner equipped with a 15-channel head array coil (Philips Healthcare, Best, the Netherlands). In conventional 3D spiral acquisitions, a stack-of-spirals (SOS) trajectory³³ (Fig. 1b) is commonly used. However, the SOS trajectory with SENSE³⁴ acceleration in the slice direction suffers SNR losses due to reduced data and regional SNR variations arising from coil sensitivity profiles, as reflected by the geometry factor (g-factor). Recently, a spiral-staircase (SSC) trajectory (Fig. 1c) has been designed.³⁰ The SSC trajectory distributes the spiral interleaves along the kz direction and periodically rotates the interleaves, with each interleaf at a unique kz position. The SSC trajectory has demonstrated higher SNR performance compared to the conventional SOS trajectory.³⁰ The SSC trajectory was incorporated into the proposed spiral method to specify the spiral interleaves in the 3D k-space.

Fig. 1.

The schematic diagram of the dual-echo spiral sequence. (a) shows the dual-echo acquisition with a spiral-out readout. (b) shows the conventional stack of spirals trajectory in 3D k-space, and (c) delineates the spiral staircase trajectory used in this work. Compared to stack of spirals, the spiral staircase trajectory is generated by setting the slice encoding gradient area to spread the spiral interleaves in the kz direction more closely in a pattern coordinated with the spiral rotation angle (as detailed in Ref. 30). When T₁ enhancement is prescribed, in addition to the spoiling gradients along the slice-selection and readout direction, RF spoiling is applied by adding additional variable phases to the RF pulses to generate incoherent signals, which help improve the T₁ contrast.

One flexibility of the SSC trajectory is the temporal distribution of the spiral interleaves. A mixed spiral interleaf ordering scheme was chosen to control the temporal order of the spiral interleaves, which combines the features of a skip ordering and a two-way ordering scheme.³⁵ In the skip ordering scheme, odd interleaves were collected first, followed by even interleaves (e.g., 1, 3, 5, 7 …, 2, 4, 6, 8, …). In the two-way scheme, the first half and the second half of the interleaves were interleaved, and the second half was also reversed (e.g., 1, N, 2, N-1, 3, N-2, …). The mixed spiral interleaf ordering scheme helps displace many of the artifacts at distant locations in the image, reducing aliasing coherence and consequently improving g-factors in the SENSE reconstruction.^30,35 The two echoes in the dual-echo spiral sequence were acquired with the same trajectories. While an accurate assessment of the impact of these ordering schemes from dynamically varying perfusion signals requires a large set of patient data, simulations have been conducted (see Supplemental Fig. S2) and demonstrated similar artifacts reduction as in Ref. 35.

DSC/DCE-MRI requires high temporal resolution to accurately track the passage of contrast agent through the capillary bed and tissues. To accelerate the scan speed, three complementary strategies were applied: i) SENSE in the slice encoding direction with a reduction factor R_SENSE = 2 (Fig. 2a); ii) a variable-density spiral readout³¹ designed to fully sample the center of k-space with a radius r = 0.3 and undersample the outer k-space with a reduction factor R_spiral = 2 (Fig. 2b); iii) a sliding-window strategy in the dynamic time series by acquiring rotated k-space and sharing undersampled high-frequency k-space data among adjacent dynamic scans (Fig. 2c). The sliding-window strategy and in-plane spiral undersampling were complementary to each other, enabled by the relative rotation of all spiral interleaves in every other dynamic acquisition by half of the angle between adjacent spiral interleaves (Fig. 2c). To reconstruct the images, the rotated spiral data were pre-processed to combine high-frequency k-space data from adjacent dynamic scans while retaining the data in the center of k-space from a single dynamic scan. This resulted in fully sampled in-plane spiral data, while preserving the image contrast with respect to the target dynamic scan (Fig. 2c). The sampling density compensation function for non-Cartesian trajectory was determined based on the combined data set. An iterative conjugate gradient algorithm was then used to reconstruct the SENSE-accelerated spiral staircase data³⁰, with an advanced deblurring technique²⁹ integrated into the iterative loop, using a pre-acquired B₀ field map. The image reconstruction was performed offline using a Graphic Programming Interface.³⁶

Fig. 2.

Approaches of acceleration used in the proposed spiral sequence, including (a) SENSE in the through-plane direction, (b) in-plane variable-density undersampling, and (c) sliding-window in the dynamic series along with combining high-frequency data from adjacent dynamics prior to spiral/SENSE reconstruction.

MRI protocol

The dual-echo spiral scan was acquired with the following imaging parameters: FOV = 224 × 224 mm², resolution = 2.5 × 2.5 mm², slice thickness = 5 mm, slice gap = 0 mm, number of slices = 20, 3D oversampling factor (slice) = 1.4, through-plane SENSE factor R_SENSE = 2, in-plane spiral undersampling factor R_spiral = 2, number of spiral interleaves = 3, spiral readout duration = 10.8 ms, TE₁ = 0.7 ms, TE₂ = 18 ms, TR = 40 ms, FA = 18°, dynamic scan time = 1.7 sec. Bloch simulations were performed with various FAs using typical T₁, T₂, and proton density values of white and gray matter and CSF at 3 T. The FA that generated good white and gray matter signals and low CSF signals was chosen as the optimal FA, which was also verified by comparing volunteer data acquired with various FAs. An FA of 18° was selected for patient studies. At our institution, a single-echo dual-dose DSC-MRI scan is included in the standard-of-care (SOC) brain tumor protocol. This technique has been validated across multiple sites and studies to reliably differentiate tumor recurrence and post-treatment radiation effects with high sensitivity and specificity (> 90%),^17,37 thereby serving as a reasonable ground truth for reference when assessing the spiral technique. The DSC data were acquired using single-echo EPI with matched slice coverage and resolution, EPI factor = 39, partial Fourier factor = 0.7, in-plane SENSE factor R_SENSE = 2.3, TE = 30 ms, TR = 1400 ms, FA = 60°, dynamic scan time = 1.4 sec. For comparison, a dual-echo EPI sequence proposed in a previous study²³ was also scanned on volunteers with similar parameters as the above single-echo EPI, while TE₁ = 7.4 ms, TE₂ = 34 ms, TR = 582 ms, FA = 75°, simultaneous multi-slice factor = 2, dynamic scan time = 0.6 sec.

Data Collection

To evaluate the 3D dual-echo spiral technique, four healthy volunteers were scanned without the administration of contrast agent. To verify the optimal FA in the spiral scans, data were acquired with a range of FAs (9–24°, with a step-size of 3°) around the FA determined through Bloch simulation (18°, see Supplemental Fig. S3). To compare the image quality with the different acceleration methods, data were acquired without/with through-plane SENSE or in-plane variable-density undersampling. To compare the signal tSNR performance of the 3D dual-echo spiral sequence with 2D dual-echo and single-echo EPI, volunteers were scanned with 25 dynamics to allow for sufficient statistical power.

To assess the performance in estimating the perfusion parameters, the spiral method was scanned on two tumor patients along with their clinical MRI exams. The 3D spiral data were acquired during the preload injection with 85 dynamics in 2 minutes and 24 seconds. The SOC 2D single-echo EPI was scanned during the second bolus injection (100 dynamics, 2 minutes and 20 seconds) after a 6-minute delay. A gadolinium-based contrast agent (gadobutrol, Gadavist) was administered in both injections at a dose of 0.1 mmol/kg and a rate of 3ml/s using a power injector after 30 seconds of baseline acquisition. The spiral sequence was scanned on one patient with RF spoiling and one patient without to examine its impact.

All volunteer and patient studies were approved by the institutional review board at St. Joseph’s Hospital and Medical Center, Phoenix, AZ. Informed consent was obtained from volunteers annually and from patients prior to the MRI scans.

Data Analysis

Volunteer data acquired with 25 dynamics using the single-echo and dual-echo EPI methods and the proposed spiral technique were processed to compute the voxel-wise signal tSNR map by dividing the mean by the temporal standard deviation. Single-echo EPI and dual-echo spiral time-courses of tumor patient data were converted to ΔR₂* curves for analysis. For dual-echo spiral data, the dynamic T₁ information was derived by extrapolating the dual-echo signals to TE = 0. Leakage correction was performed using the Boxerman-Schmainda-Weisskoff method.^38,39 An automated approach was used to determine the arterial input function.^40,41 Subsequently, rCBV values were calculated from integration of the ΔR2* time-courses and normalized by the mean value from regions of normal-appearing white matter (manually selected from relatively uniform locations near the lateral ventricles on the side contralateral to the tumor). K^trans maps were fit using a reference T₁ value (1.5 s) to the extended Toft’s model.⁴² The pixel-wise ΔR₂* tSNR was assessed by dividing the peak ΔR₂* value by the standard deviation of the steady-state baseline ΔR₂* of the patient data. Correlation and agreement between single-echo EPI and dual-echo spiral rCBV estimation were evaluated using linear Deming regression and Lin’s Concordance Correlation Coefficient (CCC). The analysis was performed using an in-house MATLAB package (version R2023a, Mathworks, MA, USA). More details of perfusion data analysis have been provided by Stokes at al.²³ Due to the small patient cohort and the technology development in nature, statistical analysis was not performed.

Results

Fig. 3 shows one representative slice acquired at various flip angles with the spiral sequence (a) and the mean signal over the whole volume from four volunteers (b). The flip angle affected both the signal intensity and the image contrast. The image acquired at a flip angle of 18° showed high white and gray matter signals and relatively dark CSF signal. The plot indicated that the mean signal peaked at approximately a flip angle of 18°, consistent with the prediction from Bloch simulation. In the rest of the study, a flip angle of 18° was used in all spiral scans.

Fig. 3.

Optimization of flip angle for the spiral sequence. (a) shows the images acquired with various flip angles: CSF appears darker compared to surrounding tissues at a flip angle of 15° or higher. (b) plots the average signal over the entire brain volume with respect to flip angles measured from four volunteers: the overall signal reaches its maximum at approximately a flip angle of 18°.

Fig. 4 qualitatively examines the image quality of the spiral results acquired with through-plane SENSE acceleration and/or in-plane variable-density undersampling with sliding-window reconstruction. Four slices from a volunteer are illustrated. The images without and with SENSE acceleration illustrated comparable image quality (Fig. 4a vs 4c, Fig. 4b vs 4d), confirming the quality of the SSC trajectory with parallel imaging as demonstrated by Anderson et al.³⁰ The quality of images without and with in-plane undersampling and sliding-window reconstruction was also comparable (Fig. 4a vs 4b, Fig. 4c vs 4d), as the latter were fully sampled after sliding-window reconstruction. The images with both through-plane SENSE and in-plane variable-density undersampling exhibited overall satisfactory image quality. Additional images are presented in Supplemental Fig. S4.

Fig. 4.

Comparison of the impact of acquisition acceleration on image quality. The images without and with SENSE (a vs. c and b vs. d, respectively) are comparable. Images without or with in-plane undersampling (a vs. b and c vs. d, respectively) are also similar. Images with combined SENSE and in-plane undersampling (d) are close to those without any acceleration (a).

Fig. 5 compares the quality of the spiral images to the EPI images in two volunteers. It was observed that the distortion and signal pile-up artifacts in the EPI images (Figs. 5c), frequently present around the frontal and temporal lobes, were substantially reduced in the spiral images (Fig. 5a). The extent of artifact reduction in the long TE image was less prominent than that in the short TE image (Fig. 5b vs Figs. 5d–5e), likely because the artifacts manifested as signal loss at long TE rather than distortion or signal pile-up due to accumulated intra-voxel dephasing.

Fig. 5.

Geometric distortion artifact reduction in spiral acquisition (a, b) compared to EPI scans (c – d). Yellow arrows point to typical distortion artifacts (signal pileup) in EPI scans in the frontal or temporal lobes. In TE₂ images (b, d) and single-echo EPI images (e), these artifacts mostly appeared as signal losses.

Fig. 6 inspects the signal tSNR performance of the spiral acquisition and the EPI scans. Fig. 6a shows the signal tSNR maps from two volunteers, and Fig. 6b plots the volumetric mean from all four volunteers. These results demonstrated the overall superior signal tSNR performance of the spiral acquisition compared to that of the dual- or single-echo EPI scans.

Fig. 6.

tSNR measurements for both spiral and EPI scans. The representative tSNR maps from two volunteers (a) demonstrated higher tSNR in the spiral results compared to the EPI scans. (b) plots the mean tSNR across the entire brain volume from four volunteers.

The quantitative perfusion measurements from patient #1 with multifocal glioblastoma are shown in Fig. 7. The leakage-corrected rCBV (LC-rCBV) maps demonstrated that the single-dose dual-echo spiral scan (Fig. 7b) achieved comparable rCBV quantification to SOC dual-dose single-echo EPI (Fig. 7a). Furthermore, the K^trans maps (Fig. 7c) can only be derived from the dual-echo spiral data. In the K^trans maps, high-grade glioblastoma in the parietal lobe showed elevated signals, as also demonstrated by the enhancing tissues in the post-contrast T₁ images (Fig. 7d). No elevated rCBV values were observed for this large enhancing tumor in the left hemisphere, which was also cross-verified using two commercial perfusion analysis software, IR Rad Tech (Imaging Biometrics, WI, USA) and OLEA (OLEA Medical, La Ciotat, France) (refer to Supplemental Fig. S5). The LC-rCBV maps across the entire brain from the spiral and EPI data are listed in Supplemental Fig. S6, allowing for inspection of the performance across the entire volume by interested readers.

Fig. 7.

Quantitative EPI-based rCBV maps (a) and spiral-based rCBV (b) and K^trans (c) maps from patient #1. All three maps (a, b, c) used the same color scale shown at the bottom. The leakage-corrected rCBV maps showed strong agreement between single-echo EPI (a) and the dual-echo spiral (b) scans. The K^trans (min⁻¹) maps (c) derived from the dual-echo spiral data revealed multifocal tumors, supported by the presence of enhancing tissues in the post-contrast T₁-weighted images (d).

Fig. 8 presents (a) the measured ΔR₂* tSNR maps and (b) the correlation between rCBV values derived from EPI and spiral data at three slices from patient #1. Spiral results demonstrated higher ΔR₂* tSNR than EPI (the whole brain average is about 28.22 for EPI and 37.66 for spiral). Lin’s CCC is 0.88, aligning with the results of a similar study²³.

Fig. 8.

The measured ΔR₂* tSNR maps (a) and the correlation plot between rCBV values derived from EPI and spiral data (b) at 3 slices from patient #1. In (b), the solid line indicates the linear Deming regression, while the dashed line shows the line of unity.

The quantitative measurements from patient #2 with glioblastoma with a primitive neuronal component are shown in Fig. 9. The spiral data were acquired without RF spoiling, unlike that from Patient #1. Consistent with the observation from patient #1, the LC-rCBV maps obtained from the spiral data (Fig. 9b) were comparable to those generated from the EPI data (Fig. 9a). The small enhancing areas in the post-contrast T₁-weighted images (Fig. 9d) next to the resection cavity did not show increased rCBV or K^trans values, suggesting that they were likely caused by post-treatment effects. In this spiral data set acquired without RF spoiling, no apparent difference was observed compared to either the EPI rCBV maps or those results from patient #1. However, more data are required to verify its effect on the resulting quantitative measurements.

Fig. 9.

Results from patient #2. The spiral data were acquired without RF spoiling. The rCBV maps exhibited similar accuracy in both EPI (a) and spiral (b) scans. The enhancing tissues in the post-contrast T₁-weighted images (d) did not exhibit increased rCBV or K^trans (min⁻¹) values, suggesting that these effects were likely due to post-treatment effects.

DISCUSSIONS

In this work, a spiral-based 3D dual-echo sequence with a single dose for simultaneous DSC- and DCE-MRI was developed and compared to a SOC 2D dual-dose single-echo EPI technique and a recently studied 2D dual-echo EPI method. One major benefit over the current SOC single-echo EPI with dual doses is the reduction of contrast dose by 50%. In the clinic, the reduction of contrast dose has an immediate impact on patient care, including acute adverse reactions, nephrogenic systemic fibrosis in patients with impaired kidney functions and long-term gadolinium retention.⁴³ Another advantage of spiral acquisitions over EPI acquisitions is the reduced signal distortions in areas with strong field inhomogeneities due to susceptibility effects at tissue interfaces, around surgical cavities, or near metal implants. The degree of improvement is affected by many factors such as the extent of the field inhomogeneity, the duration of the spiral waveform, the echo time, and so on. As evidenced in Fig. 5, the distortion artifacts are significantly reduced in the first-echo image of the spiral scan, compared to the first-echo image of the dual-echo EPI scan and the image from the SOC single-echo EPI scan as well. In the second-echo images, the distortion artifacts are mostly manifested as signal loss due to accumulated intra-voxel dephasing. In these second-echo images, the spiral results are in general better than the EPI images, while the improvement is not as large as that in the first-echo image. Another improvement demonstrated through the volunteer study is the enhanced signal tSNR of the spiral images. This increased signal tSNR is collectively contributed to by a variety of factors, such as scan trajectory (3D vs. 2D, spiral vs. EPI), TE (T₂* decay), TR (T₁ relaxation), acceleration and related g-factor, flip angle, etc. One benefit of the dual-echo spiral technique over the dual-echo EPI technique is the preferable imaging parameters. The 2D dual-echo EPI results in a TR of about 600 ms, which reduces the sensitivity to T₁ relaxation.²³ In contrast, the 3D spiral scan has a TR of approximately 40 ms, making it much more sensitive to T₁ relaxation effect.

The proposed 3D dual-echo spiral sequence achieved a temporal resolution of 1.7 seconds, slightly longer than that of the SOC single-echo EPI scan (1.4 seconds), with the same imaging coverage and spatial resolution. For comparison, the temporal resolution was 1.35 seconds for 13 slices with a 2D dual-echo spiral sequence,²⁶ about 1.5 seconds for 15 slices with a 2D spin and gradient echo spiral sequence,²⁷ and would increase to about 2 seconds for a 20-slice coverage, although simultaneous multi-slice acquisition can improve the temporal resolution of these 2D scans albeit at the cost of SNR performance. Long temporal resolution may slightly affect the CBF estimation, while having little impact on the CBV accuracy.⁴⁴ Although a temporal resolution of 1.5 seconds or shorter is commonly used,^19,44 a temporal resolution of 1.8 seconds has been used and demonstrated good results in the combined spin and gradient echo sequences.^45,46 There were several factors and considerations that affected the temporal resolution in the proposed dual-echo spiral technique. In this study, we used a wide-bore gradient system, which typically has moderate gradient performance (max gradient strength: 20 mT/m, max slew rate: 140 mT/m/ms). Gradient performance affects the overall efficiency of the spiral trajectory. With a nominal undersampling factor of R_spiral = 2 and a spiral readout duration approximately 10 ms at 3 T, the variable-density spiral trajectory design in this study reduced the number of spiral interleaves to 3, given that the center of k-space was fully sampled and there was a transition zone between the center and outer k-space. This resulted in a “true” acceleration factor of ~1.67. Higher nominal acceleration factors, e.g., 3, could be implemented, but there was only marginal gain in the resulting “true” acceleration factor due to the limiting gradient performance and the small number of interleaves needed to fill the k-space (therefore this was not used in this study). Longer spiral readouts can also help reduce the temporal resolution, at the cost of residual blurring, while the improvement is again limited by the small number of interleaves. Another factor was the acceleration factor in the slice direction. With SSC, an acceleration factor of R_SENSE = 3 could be used but was not chosen in this study. One consideration was the slightly increased artifact level, as illustrated by Anderson et al.;³⁰ the other was to preserve good SNR performance. Given the limited number of patients enrolled to verify the feasibility of this technique, only R_SENSE = 2 was tested. In this work, TE₂ was chosen based on considerations such as its impact on TR and T₂* decay. TE₂ and ΔTE significantly contribute to the quality of the perfusion maps. A spiral-out spiral-in scheme helps increase TE₂ and ΔTE, which was not investigated in this work due to concerns over the different impact on the two echoes from factors such as trajectory fidelity and eddy currents.⁴⁷ A comprehensive and systematic optimization of these parameters is feasible but may require a large patient cohort. This will be investigated in a future patient study.

The concept of using a spiral dual-echo technique for perfusion imaging has been previously explored.^26,28 This proposed method includes several non-trivial additions, such as a 3D scan, an SSC trajectory with improved parallel imaging performance, a variable-density trajectory design for in-plane undersampling, a sliding-window approach for increased temporal resolution, an enhanced deblurring technique, and a conjugate gradient algorithm for joint parallel imaging reconstruction and spiral deblurring. All these components collectively helped improve the overall performance of the spiral technique. However, we did not directly compare our 3D spiral scan with a 2D spiral scan for DSC/DCE, as both need to be acquired during the first contrast injection.

The quantitative rCBV maps from spiral data seemed slightly less sharp than those from EPI data. There are multiple possible factors that can affect the sharpness of the spiral-based maps. The EPI images were reconstructed online with vendor’s specific filters, while the spiral images were generated offline and further processed with a diffusion filter,^48–50 which may not match the vendor filtering. Furthermore, the voxel size in the spiral data is relatively large compared to a typical anatomical scan using 3D spiral acquisition; the intra-voxel B₀ variation in this large voxel may not be reflected in the B₀ field map and consequently affects the deblurring performance. Another factor is the area of the in-plane k-space coverage of the spiral acquisition, which is smaller than that of the Cartesian EPI data and slightly affects the spiral image sharpness. Increasing the k-space coverage and consequently improving the image sharpness is possible but was not used in this study, primarily to retain high temporal resolution. The difference in image sharpness was not easily discernible in the source images, possibly because of the low contrast or the different contrast between the spiral and EPI images. Nonetheless, the sharpness of the spiral-based rCBV maps generated from our in-house MATLAB package seems comparable to the EPI-based maps computed with one commercial software (OLEA, as shown in Supplemental Fig. S5), implying that post-processing filtering might be a contributing factor.

The reconstruction code was implemented with a combination of Python and C languages. The total offline reconstruction time was approximately 22 minutes, partially employing multi-threading for C-based subroutines. The reconstruction time is expected to be significantly shorter if the functions were fully implemented in C and run on a host computer with more CPU cores and/or a GPU, and therefore not be a barrier to clinical translation.

In this study, one limitation is that we were only able to directly compare the proposed 3D dual-echo spiral technique with the previously proposed 2D dual-echo EPI technique on volunteers without contrast agent, as both need to be run during the preload contrast injection when scanning patients; as a consequence, we were not able to directly compare the accuracy of the resulting rCBV and K^trans maps between the spiral acquisition and the dual-echo EPI method. Fortunately, both the previous patient study by Stokes et al.²³ and the patient study in this work demonstrated that a dual-echo acquisition scheme with a single dose (either EPI or spiral) achieved quantitative DSC metrics comparable to those from the SOC single-echo dual-dose EPI, with additional benefits such as the simultaneous estimation of DCE information and the reduction of contrast agent dose. In addition, the spiral results from healthy volunteer scans illustrated improved artifact reduction and signal tSNR performance compared to the dual-echo EPI technique. Together, these indirect and direct comparisons demonstrated the advantages of the proposed dual-echo spiral technique over the dual-echo EPI method. Another limitation is that we only scanned two patients, primarily for verifying the feasibility of the spiral sequence. The next step will focus on collecting data from a larger cohort of patients to statistically compare the proposed spiral sequence to the SOC single-echo EPI technique.

CONCLUSION

In summary, a single-dose 3D dual-echo spiral pulse sequence has been developed and exhibited reduced geometric distortion artifacts and improved tSNR compared to conventional EPI. It allows for simultaneous and accurate quantification of DSC and DCE perfusion metrics, providing a promising and reliable alternative to EPI-based DSC- and DCE-MRI for improved characterization of tumor status, therapy response assessment, and clinical trial use.

Supplementary Material

Supinfo

Supplemental Fig. S1. Simulated maximum measurable contrast agent concentration (Ref 32), C_max. Assuming contrast agent relaxivities of 3.3 and 87 for r₁ and r₂*, respectively, and a pre-contrast R₁ of 1.5 s⁻¹. The TE₁/TE₂/TR are 1/18/40 ms (3D spiral), 1/18/700 ms (2D spiral), 8/30/1500 ms (2D EPI), respectively. It is observed that C_max is 0.52 mM and 2.80 mM for 2D and 3D spiral acquisition, respectively.

Supplemental Fig. S2. Simulation of the impact of the ordering schemes from the dynamically changing perfusion signals: (a) linear, (b) skip, (c) two-way, and (d) mixed. The simulation was based on the 3D SSC acquisition with a matrix size of 128 × 128, 6 spiral interleaves, disc size of 6 × 6. The signal changes temporally from 0 to 1 throughout all interleaves in the 3D k-space data set. In this protocol, the skip and mixed ordering schemes spread the intense swirl-like artifact to the edge of the image, similar to that demonstrated in Ref. 35.

Supplemental Fig. S3. Bloch simulation of signals vs flip angles. The proton density/T₁/T₂ values used for WM, GM, and CSF are 0.65/800ms/90ms, 0.75/1200ms/100ms, 1.0/3300ms/2100ms, respectively. T₁ reduction in enhancing tumors after contrast injection shifts the signal peak slightly towards larger flip angles, while the signals vary slowly around the peak (not shown). Therefore, it does not significantly affect the selection of a “target” flip angle.

Supplemental Fig. S4. Additional volunteer images demonstrating image quality with various acquisition acceleration methods.

Supplemental Fig. S5. rCBV maps generated from EPI data using an in-house MATLAB package, IB Rad Tech, and OLEA, respectively, as well as from spiral data using an in-house MATLAB package. The resulting rCBV maps were comparable, with no elevated rCBV values were observed for the large enhancing tumor in the left lobe.

Supplemental Fig. S6. rCBV maps of the entire brain generated from EPI and spiral scans of patient #1. Masks were generated from the average of dynamic 3 to dynamic 8 (baselines) of EPI images for EPI-based rCBV maps and from the average of dynamic 3 to dynamic 8 of the mean of TE₁ and TE₂ spiral images for spiral-based rCBV maps. These rCBV maps were then overlaid onto the corresponding average EPI and spiral images.