Non‐Subtractive Arterial Spin Labeling‐Based (NSASL) Renal Magnetic Resonance Angiography (MRA): Development and Clinical Feasibility Evaluation

ABSTRACT

Background

Non‐contrast renal MR angiography (MRA) is valuable for patients who cannot receive contrast agents or when avoiding radiation is desired. However, the conventional inflow inversion recovery (IFIR) method is limited by incomplete background suppression, venous contamination, and motion sensitivity.

Purpose

To develop and evaluate a non‐subtractive arterial spin labeling‐based (NSASL) sequence for renal MRA and compare it with IFIR in healthy volunteers, with exploratory feasibility assessment relative to contrast‐enhanced (CE)‐MRA or computed tomography angiography (CTA) in patients.

Study Type

Prospective.

Subjects

Thirty healthy volunteers (10 male, 35.6 $\pm$ 14.0 years) and six participants with renal diseases or high blood pressure (2 male, 62.7 $\pm$ 9.03 years).

Field Strength/Sequence

1.5 T; 3D stack‐of‐stars balanced steady‐state free precession (bSSFP) NSASL MRA, 3D Cartesian bSSFP IFIR MRA, and CE‐MRA.

Assessment

Three radiologists independently rated image quality (main and branch renal arteries visualization, motion artifacts, vessel‐to‐background contrast, diagnostic confidence) on a 5‐point scale. SNR efficiency (or SNR and time) and contrast ratio (CR) were also measured.

Statistical Tests

Intraclass coefficient (ICC), Shapiro–Wilk's test, paired Student's t‐test, and Wilcoxon signed‐rank test with Bonferroni correction, with p < 0.05 considered statistically significant.

Results

ICC was good to excellent (ICC = 0.61–0.89) for average subjective scores. Compared with IFIR, NSASL showed significantly better vessel‐to‐background contrast (approximately fivefold increase in CR, Cohen's |d| = 2.54; +1 subjective score, |r| = 0.88), improved renal arteries visualization (+0.7 points, |r| = 0.67, corrected p   $\le$0.003), fewer motion artifacts (+0.6 points, |r| = 0.67, corrected p = 0.002), and higher diagnostic confidence (+0.6 points, |r| = 0.80, corrected p < 0.001). Acquisition time was reduced from 267.3 ± 69.0 s to 240.2 ± 51.3 s (Cohen's |d| = 0.54, corrected p = 0.018) while SNR efficiency was moderately lower (~26%, Cohen's |d| = 1.99, corrected p = 0.002). In participants with disease, NSASL yielded similar diagnostic confidence to CE‐MRA (n = 4, +0.7 points, p = 0.194) and to CTA (n = 2, −0.2 points, p = 0.317).

Data Conclusion

NSASL significantly outperformed IFIR, with improved background suppression, vessel conspicuity, motion tolerance, and scan time in healthy volunteers.

Evidence Level

2.

Technical Efficacy

Stage 2.

Plain Language Summary

This study developed a safer way to image kidney arteries using MRI without contrast agents or radiation. The new method was designed to strongly suppress background tissues while preserving clear signals from blood flow. It was tested in healthy volunteers and compared with a commonly used non‐contrast MRI technique. The new approach produced clearer vessel images, better background suppression, and shorter scan times. Early tests in a small group of patients also showed promising feasibility. These findings suggest that this method may offer a more effective and entirely non‐invasive option for evaluating kidney blood vessels.

1. Introduction

Accurate visualization of renal arteries is important for the diagnosis and management of conditions such as renal artery stenosis, embolism, aneurysm, tumors, and vascular insufficiency [1]. Contrast‐enhanced (CE) computed tomography angiography (CTA) and magnetic resonance angiography (MRA) are widely used non‐invasive clinical techniques for this purpose [2, 3]. While CTA provides high spatial resolution, it exposes patients to ionizing radiation and iodine‐based contrast agents; the latter can cause allergic reactions and pose risks to renal function [4]. ce‐MRA avoids radiation; yet the administration of gadolinium‐based contrast agents carries risks of nephrogenic systemic fibrosis and potential renal toxicity, particularly in patients with impaired kidney function [5, 6].

These limitations have driven increasing interest in non‐contrast (NC) MRA methods, which mitigate safety risks and reduce costs, making them well‐suited for screening and longitudinal follow‐up [7]. However, existing NC MRA approaches have important drawbacks. Time‐of‐flight (TOF) MRA is susceptible to saturation effects, leading to in‐plane signal loss and stenosis overestimation, limiting its reliability in complex and tortuous renal arteries [8, 9]. Phase contrast (PC) MRA provides additional quantitative flow information, but its reliance on subtraction increases acquisition time and motion sensitivity, limiting routine clinical use [10, 11, 12]. Balanced steady‐state free precession (bSSFP) combined with inflow inversion recovery (IFIR) is a well‐established commercial NC MRA technique in clinical practice, benefitting from strong blood‐tissue contrasts and intrinsic flow compensation [13, 14]. Yet, it relies on a single inversion pulse, and when arterial transit times (ATT) are prolonged, background tissues recover beyond the null point, thereby diminishing vessel conspicuity [15, 16]. Arterial spin labeling (ASL), particularly pseudo continuous ASL (pCASL) combined with bSSFP, has shown promising results by labeling inflowing blood as an endogenous tracer. However, conventional ASL methods require label‐control subtraction, resulting in prolonged scanning time, lower SNR, and increased motion sensitivity, limiting broader clinical translation [17, 18, 19].

To address these challenges, a novel non‐subtractive ASL‐based (NSASL) MRA technique was developed. This method integrated a multi‐inversion strategy with interleaved pCASL labeling to decouple background suppression (inversion time) from ATT (inflowing blood), enabling near‐complete background suppression while preserving sufficient arterial inflow signal. In addition, respiratory triggering, radial stack‐of‐stars (SoS) sampling, and a newly designed trajectory switching mode were incorporated to further reduce motion artifacts and stabilize central‐out bSSFP acquisition, a strategy previously considered impractical but made feasible through these technical improvements. Together, these innovations aimed to enhance arterial conspicuity while overcoming the inherent limitations of conventional subtraction‐based methods.

This study aimed to assess the feasibility of the proposed new NSASL sequence compared with the established IFIR MRA technique in healthy volunteers and to preliminarily explore its clinical feasibility in relation to CE‐MRA and CTA.

2. Materials and Methods

2.1. Study Population

This prospective study was approved by the local Institutional Review Board of Ningbo Number 2 Hospital, and written informed consent was obtained from all participants prior to enrollment.

For the technical feasibility evaluation, healthy adults were recruited between June and July 2025. Individuals with a history of kidney diseases, renal abnormalities, prior renal surgery, or pregnancy were excluded. Demographic information (age, height, and weight) and physiological parameters, including systolic blood pressure (SBP), diastolic blood pressure (DBP), heartbeat rate (HBR) before and after scanning, and respiratory rate (RR) during scanning, were recorded for all participants.

To explore clinical feasibility rather than diagnostic performance, a small cohort of participants who were clinically referred to CE‐MRA or CTA for various indications, but not specifically for renal vascular diseases, were prospectively recruited between July and August 2025.

2.2. Sequence Design

As illustrated in Figure 1a, the conventional IFIR MRA technique applied a single inversion pulse and acquired data as background signals recovered through the null point. In contrast, the proposed NSASL MRA sequence incorporated a preparation module that combined interleaved pCASL blood labeling with multiple inversion background suppression, achieving near‐complete background nulling while preserving the majority of arterial inflow signals prior to acquisition. Each acquisition block was initiated by a respiratory trigger, followed by a spatially selective saturation pulse to suppress the longitudinal magnetization of static tissues and venous blood.

FIGURE 1.

(a) Renal MRA setup and inversion RF pulse locations of IFIR sequence. (b) Timing diagrams of IFIR sequence with the longest ATT defined as TS. (c) Renal MRA setup, saturation, inversion, and pCASL RF pulse locations of NSASL sequence. (d) Timing diagram of NSASL sequence with the longest ATT defined as TS. (e) Longitudinal signal evolution of arterial blood flowing into the imaging region at different time points (A to E), compared with different static background tissues during NSASL labeling period. (f) Arterial blood longitudinal signal with different ATT values at the imaging point (time = 0).

Using the longest expected ATT as the labeling time (TS, defined as the interval between saturation and central k‐space acquisition), multiple spatially selective adiabatic inversion pulses were applied at carefully optimized intervals to maintain the lowest background tissue signal based on the algorithm proposed by Maleki et al. [20] This strategy enabled suppression of background tissues with a broad range of T1 values (250 to 4200 ms) to less than 1% at the central k‐space acquisition point (time = 0). In this study, four inversion pulses were used. The saturation and inversion pulses covered the imaging volume and extended distally to suppress both static background tissues and venous blood.

pCASL labeling modules were inserted into the time segments immediately preceding the odd‐numbered inversions, ensuring that the inflowing arterial blood was prepared and preserved with a positive signal at acquisition (Figure 1d). The pCASL labeling plane was positioned at the proximal edge of the imaging slab.

A bSSFP readout was adopted due to its intrinsic T2/T1 contrast, which is well‐suited for bright‐blood imaging [21]. To improve motion robustness and enable compressed sensing (CS) acceleration, a SoS radial sampling trajectory was employed with data acquired in a central‐out order along the partition (slice) encoding direction to preserve vascular contrast.

However, conventional SoS golden angle rotation introduced large spoke‐to‐spoke gradient jumps, resulting in non‐constant eddy currents and bSSFP signal distortion [22]. In addition, bSSFP required prolonged transition scans to stabilize oscillations, and its inherent bright fat signal can amplify fluctuations, making central‐out strategies rarely adopted in practice.

To address these issues, a small golden remainder angle switching mode was developed (Figure 2). In this scheme, k‐space was filled using a central‐out partition‐encoded (k _z ) trajectory, acquired in an interleaved fashion after each respiratory trigger. An even number of M successive spokes were acquired at the same partition, each rotated by a tiny angular increment to ensure non‐overlapping k‐space coverage while minimizing gradient variation [23]. This design provided two advantages: (1) it yielded uniform k‐space coverage, thereby reducing streaking artifacts and (2) interleaving M partition‐encoding spokes allowed the polarity‐alternative RF pulses to average out bSSFP signal oscillations, thereby improving stability. The angular increment ${\phi }_N$ was calculated as the remainder of the standard golden angle series divided by $\pi$ as shown below, where N is a Fibonacci sequence member and ${\psi }_1$ is the commonly used golden angle of 111.25° [24].

$${\phi }_N=\mathit{mod}\left({\psi }_{1}N,\pi \right)$$

FIGURE 2.

Sampling trajectory design. (a) SoS transverse plane distributions of a 10‐spoke schematic example using the tiny golden angle and the small golden remainder angle (M = 2). (b) Within each partition, $M$ spokes were collected using a small golden remainder angle switching strategy, with alternating rotating direction between adjacent partitions. Partition encoding was performed in a central‐out order in each trigger. For each new trigger group, the initial spoke angle was offset by ${\psi }_1$. The acquisition order of the trajectory was indicated by dashed arrows, with darker‐colored spokes denoting earlier acquisitions.

To further reduce phase accumulation, the rotation direction of the tiny‐angle increments was alternated between partitions (Figure 2b). The initial spoke angle following each respiratory trigger was rotated by a full golden angle ${\psi }_1$ to ensure rapid k‐space coverage, while an additional cumulative angle offset of $\left(M-1\right){\phi }_N$ was applied every N triggers. Compared with conventional sequential tiny golden angle rotations (${\psi }_N=\frac{\pi }{\frac{1+\sqrt{5}}{2}+N-1}$), this strategy achieved faster uniform k‐space coverage and reduced phase errors (Figure 2a) [25, 26]. In this study, we selected N = 34, M = 4, and ${\phi }_N$ = 2.35°.

Additionally, to suppress fat and further stabilize the transient‐state signal, periodic spectral presaturation with inversion recovery (SPIR) fat suppressions were inserted between fast interrupted steady‐state (FISS) modules, structured as [$\alpha$/2, $\alpha$, (readout, $\alpha$) _n , $\alpha$/2, spoil] [27, 28].

2.3. Imaging Protocol

The proposed 3D NSASL MRA sequence was implemented on a 1.5 T superconductive MRI system (XGY‐Spin MRI‐R001, Xingaoyi Medical Equipment Co. Ltd., Ningbo, China) equipped with a 16‐channel phased array abdominal coil. Each labeling‐acquisition block was respiratory‐triggered using a belt transducer. Following the five preparation RF pulses at time points −1400.0, −1186.8, −741.4, −337.7, and −84.3 ms, respectively, data were acquired in the axial orientation using a central‐out SoS bSSFP readout with TE tuned to place the bSSFP stopband at the fat resonance frequency [27]. A pattern of four interleaved spokes was used, with a 2.35° angular rotation increment between adjacent spokes (Table 1).

TABLE 1.

NSASL and IFIR MRA sequence parameters.

Parameter	NSASL	IFIR
Labeling time	1400 ms	1400 ms
FOV	333 × 333 × 94 mm3	333 × 250 × 94 mm3
Filp angle	90°	90°
TR/TE	4.34/2.17 ms	3.32/1.66 ms
Bandwidth	200 kHz	200 kHz
Sampling strategy	Central‐out SoS	Sequential Cartesian
Acceleration	k z partial Fourier (0.67) Radial under‐sampling (R = 2)	k y GRAPPA (R = 2, ACS = 24)
Total under‐sampling	3	1.8
Acquisition matrix	256 sample × 200 rotation × 48 k z	256 k x  × 108 k y  × 72 k z
Acquisitions/trigger	192 (48 k z  × 4 ${\phi }_{34}$ rotation)	108 (k y )
Acquisition window	1122 ms	369 ms
Minimum trigger interval	2536 ms	1618 ms
Trigger number	50 (${\psi }_1$ rotation)	72 (k z )
Intra‐group FS	Every 16 acquisitions	No

Abbreviations: ACS, auto‐calibration signal; FOV, field of view; FS, fat suppression; IFIR, inflow inversion recovery; NSASL, non‐subtractive arterial spin labeling‐based; R, reduction factor; TE, echo time; TR, repetition time.

These undersampled k‐space signals were reconstructed using nonuniform fast Fourier transform (NUFFT) combined with CS algorithm in Matlab (2024b, Mathworks, Natick, USA) on a workstation (intel i7‐13700KF, 32 GB RAM, NVIDIA RTX 4070 Ti SUPER with 16GB VRAM). The CS regulation factor $\lambda$ was selected to be 1% of the maximum image pixel value as a trade‐off between small blood vessel blurring and image noise increase (Figure S1). The MATLAB‐based reconstruction required roughly 14 min, which was impractical for routine clinical use. To overcome this bottleneck, a GPU‐accelerated reconstruction pipeline was developed and tested on the same workstation, completing reconstructions in well under 3 min [29]. This substantial acceleration enabled near‐real‐time processing, transforming NSASL from a research prototype into a clinically viable method. Reconstructed images were zero‐padded to 512 × 512 × 144, yielding an isotropic resolution of 0.65 mm. Multi‐view maximum intensity projection (MIP) from cropped images was generated for radiologist evaluation.

For comparison, the conventional 3D IFIR MRA sequence was performed on a 1.5 T MRI system (Magnetom Altea, Siemens Healthineers, Erlangen, Germany) with an 18‐channel phased array abdominal coil. This sequence was respiratory‐triggered by Siemens integrated non‐contact sensors and used the same labeling delay (1400 ms). Images were reconstructed using MR software (syngo xA20, Siemens Healthineers, Erlangen, Germany) and zero‐padded to 512 × 384 × 144 to match the 0.65 mm isotropic resolution of NSASL. Multi‐view MIP images from cropped images were also generated in Matlab 2024b.

For exploratory feasibility assessment in patients, standard renal CE‐MRA was performed on the same Siemens 1.5 T system using a coronal breath‐hold FLASH sequence with gadolinium‐based contrast injection, while renal CTA was acquired on a 128‐slice Siemens scanner using routine abdominal angiography protocols with iodine contrast. Detailed acquisition parameters were listed in CE‐MRA and CTA imaging protocol of the Supporting Information.

All participants underwent NSASL MRA and IFIR MRA in randomized order during the same session. Total acquisition times were recorded for both techniques. Patients additionally underwent CE‐MRA or CTA after completion of the two NC MRA scans on the same day.

2.4. Sequence Module Comparison

To enable direct comparison under matched acquisition conditions, IFIR and NSASL were evaluated using the same number of respiratory triggers and identical accelerations (0.75 partial Fourier) on the aforementioned 1.5 T MRI systems of Siemens and Xingaoyi, respectively. To isolate the effects of individual sequence components, fully sampled datasets without CS were acquired in a healthy volunteer using different combinations of labeling strategies (IFIR vs. NSASL), k‐space trajectories (Cartesian vs. SoS), and acquisition orders (sequential vs. central‐out) on the same Xingaoyi 1.5 T system. Furthermore, both NSASL and IFIR were acquired with varying labeling delays (TS = 800–1600 ms) to assess sensitivity to ATT.

2.5. Image Evaluation

Reconstructed 3D images of healthy volunteers and six participants with diseases and their view‐angle‐adjustable MIPs were uploaded to the same PACS (eWorld v4, Mingtian Yiwang Technology Co. Ltd., Ningbo, China). Scanner models, sequence types, and participant identifiers were anonymized before analysis.

Objective image quality metrics included:

1. Signal‐to‐noise ratio (SNR) efficiency: the mean signal intensity (SI) of the main renal artery of interest (ROI) divided by the standard deviation (SD) of background noise, measured from four square ROIs placed in air, and the square root of total acquisition time (in seconds). For patient comparison with CTA, the SNR and scan time instead of SNR efficiency were used.

   $$\mathit{SNR}\mathit{\text{efficiency}}=\frac{{\mathit{SI}}_{\mathit{\text{renal artery}}}}{{\mathit{SD}}_{\mathit{\text{background noise}}}\times \sqrt{\mathit{\text{Time}}}}$$
2. Contrast ratio (CR): relative mean SI difference between the main renal artery ROI and renal parenchymal ROI

   $$\mathit{CR}=\frac{{\mathit{SI}}_{\mathit{\text{renal artery}}}-{\mathit{SI}}_{\mathit{\text{renal}}\text{parenchyma}}}{{\mathit{SI}}_{\mathit{\text{renal}}\text{parenchyma}}}$$

Subjective image quality was assessed independently by three blinded radiologists (K. Y., Q. D., and L. G., each specializing in abdominal imaging, with 24, 12, and 9 years of experience, respectively), who reviewed the anonymized datasets. Image quality was rated using a 5‐point scale across five categories as detailed in Table S1.

2.6. Statistical Analysis

All statistical analyses were performed using SPSS Statistics (v29, IBM Corp., New York, USA). The chi‐square test was conducted for the gender distribution evaluation of healthy volunteers. Inter‐rater reliability (IRR) among three radiologists was assessed using intra‐class correlation coefficient (ICC), from a two‐way mixed effects model with a consistency definition. ICC values were interpreted as follows: ≥ 0.75, excellent; ≥ 0.6, good; ≥ 0.4, fair; and < 0.4, poor agreement. Normality of objective measures (SNR efficiency, CR, and acquisition time) were verified with the Shapiro–Wilk's test, and all satisfied with the assumption. Paired comparisons of the objective measures were conducted using a two‐tailed Student's t‐test, and the results were reported as mean ± SD. For ordinal data of subjective image quality scoring, differences between NSASL and IFIR MRA were evaluated using the Wilcoxon signed‐rank test, reported as median (IQR, interquartile range). Statistical significance was defined as p < 0.05 with Bonferroni multiple comparisons correction applied.

3. Results

3.1. Participant Characteristics

Thirty‐one healthy volunteers were initially recruited. One male volunteer was excluded due to failed IFIR respiratory triggering, resulting in a final cohort of 30 participants (10 male). Baseline physiological parameters (HBR, SBP, and DBP) remained stable before and after scanning (Table 2), confirming good tolerance of the examinations. Although the cohort had more women than men, the difference in gender distribution was not statistically significant (p = 0.07). In addition, a small exploratory clinical cohort of six participants with diseases (2 male) was included (Table S2).

TABLE 2.

Demographic characteristics of healthy volunteers (n = 30).

Characteristics	Data	p‐value (pre‐post scan difference)
Sex (male/female)	10/20	N/A
Age (mean/range)	35.6 $\pm$ 14.0/20–72 years	N/A
	Male: 25.4 $\pm$ 3.1/20–29 years
	Female: 40.8 $\pm$ 14.5/20–72 years
BMI (mean/range)	23.0 $\pm$ 2.6 kg m−2/17.6–28.0 kg m−2	N/A
	Male: 23.4 $\pm$ 1.6/21.6–25.8 kg m−2
	Female: 22.8 $\pm$ 3.0/17.6–28.0 kg m−2
RR during scanning	16.1 $\pm$ 5.0 breaths per minute	N/A
HBR before scanning	77.6 $\pm$ 9.6 beats per minute	0.15
HBR after scanning	75.9 $\pm$ 9.7 beats per minute
SBP before scanning	116.5 $\pm$ 14.2 mmHg	0.67
SBP after scanning	115.8 $\pm$ 15.0 mmHg
DBP before scanning	76.4 $\pm$ 7.2 mmHg	0.91
DBP after scanning	76.3 $\pm$ 8.5 mmHg

Abbreviations: BMI, body mass index; DBP, diastolic blood pressure; HBR, heartbeat rate; N/A, not applicable; RR, respiratory rate; SBP, systolic blood pressure.

3.2. Sequence Module Comparison

Under matched trigger numbers and partial Fourier acceleration, NSASL achieved SNR comparable to IFIR when partial Fourier was applied along k _y , and higher SNR and CR when applied along k _z at the cost of longer acquisition time due to increased triggers (Figure S2; Tables S3, S4).

Among fully sampled acquisitions without CS, NSASL combined with central‐out SoS sampling yielded the best vessel conspicuity and background suppression compared with other combinations (Figure S3; Tables S5, S6).

When labeling delay was varied, IFIR demonstrated optimal performance near 1200 ms TS, whereas NSASL maintained stable arterial signal and effective background suppression across the wide TS range (800–1600 ms) (Figure S4).

3.3. Image Quality Evaluation

Representative images of healthy volunteers with corresponding average radiologist ratings are shown in Figure 3. NSASL consistently demonstrated superior background suppression, yielding excellent vessel conspicuity and blood‐to‐tissue contrast (Figure 3a). Veins that were frequently visible on IFIR images (Figure 3b–d) were largely reduced with NSASL. Similarly, bright signals from short‐T1 tissues (e.g., bile), often seen on IFIR, were effectively removed in NSASL (Figure 3b), although residual signals from intestinal tissues occasionally remain visible in certain NSASL cases (Figure 3c,d). NSASL demonstrated greater uniformity and continuity for visualization of renal arteries than IFIR, potentially reducing misinterpretation of stenosis or occlusion. In rare cases, IFIR produced slightly higher image quality, typically in subjects with very regular respiration that favored stable triggering (Figure 3d).

FIGURE 3.

MIP images of the renal arteries obtained with IFIR and NSASL MRA in transversal, sagittal, and coronal planes, with corresponding averaged subjective scores. (a) A 25‐year‐old man (BMI: 21.6 kg m⁻², RR: 12 breaths per minute) with the highest NSASL score. (b) A 29‐year‐old woman (BMI: 22.7 kg m⁻², RR: 28 breaths per minute) with the median NSASL score. (c) A 26‐year‐old man (BMI: 23.9 kg m⁻², RR: 21 breaths per minute) with the lowest NSASL score. (d) A 30‐year‐old woman (BMI: 21.2 kg m⁻², RR: 10 breaths per minute) with the highest IFIR score.

As presented in Figure 4 and Table 3, objective metrics demonstrated significant differences between NSASL and IFIR, even after Bonferroni correction. NSASL reduced acquisition time by ~10% (corrected p = 0.018, Cohen's |d| = 0.54), indicating greater efficiency. Although SNR efficiency exhibited a ~26% reduction (corrected p = 0.002, Cohen's |d| = 1.99), it achieved nearly an almost fivefold higher CR (corrected p < 0.001, Cohen's |d| = 2.54), representing a very large effect size. Overall, NSASL traded modest SNR loss for substantially improved vessel‐to‐background contrast and slightly shorter acquisition time, with CR improvement emerging as its most clinically relevant advantage.

FIGURE 4.

Boxplot comparing IFIR and NSASL images in terms of (a) SNR efficiency, (b) CR, and (c) subjective scores across five image quality categories. *, **, and ***indicate a Bonferroni corrected p‐value < 0.05, < 0.01, and < 0.001, respectively.

TABLE 3.

Objective comparison of NSASL and IFIR MRA image quality.

Category	IFIR	NSASL	t	Cohen's |d|	Raw p‐value	Corrected p‐value
SNR efficiency	5.0 $\pm$ 1.6	3.7 $\pm$ 1.5	3.58	1.99	0.001	0.002
CR	1.8 $\pm$ 0.5	8.5 $\pm$ 2.6	−13.89	2.54	< 0.001	< 0.001

Note: Bolded numbers indicate statistically significant p‐values (p < 0.05).

Abbreviations: CR, contrast ratio; IFIR, inflow inversion recovery; NSASL, non‐subtractive arterial spin labeling‐based; SNR, signal‐to‐noise ratio.

IRR results are summarized in Table 4. All ICC values were significant. For a single rater, ICC (3, 1) ranged from 0.34 to 0.73, indicating mostly fair to good agreement. Averaging across three raters improved consistency, yielding ICC (3, k) values of 0.61 to 0.89, which confirmed that mean scores provided sufficient reliability for subsequent analysis.

TABLE 4.

ICC for NSASL and IFIR image quality based on a two‐way mixed‐effects consistency model.

Sequence	Category	ICC (3, 1)			ICC (3, k)
		Value	95% CI	Interpretation	Value	95% CI	Interpretation
IFIR	Main renal artery	0.50	[0.28, 0.69]	Fair	0.75	[0.53, 0.87]	Excellent
	Branch renal artery	0.72	[0.55, 0.84]	Good	0.88	[0.79, 0.94]	Excellent
	Motion artifacts	0.65	[0.47, 0.80]	Good	0.85	[0.72, 0.92]	Excellent
	Vessel contrast	0.49	[0.27, 0.69]	Fair	0.74	[0.52, 0.87]	Good
	Diagnosis confidence	0.60	[0.40, 0.77]	Good	0.82	[0.67, 0.91]	Excellent
NSASL	Main renal artery	0.34	[0.11, 0.57]	Poor	0.61	[0.28, 0.80]	Fair
	Branch renal artery	0.58	[0.38, 0.75]	Fair	0.81	[0.65, 0.90]	Excellent
	Motion artifacts	0.73	[0.57, 0.85]	Good	0.89	[0.80, 0.95]	Excellent
	Vessel contrast	0.40	[0.18, 0.62]	Fair	0.67	[0.39, 0.83]	Good
	Diagnosis confidence	0.67	[0.48, 0.81]	Good	0.86	[0.74, 0.93]	Excellent

Abbreviations: CI, confidence interval; ICC, intraclass correlation coefficient; IFIR, inflow inversion recovery; NSASL, non‐subtractive arterial spin labeling‐based.

As shown in Figure 4 and Table 5, average radiologist scores demonstrated that NSASL significantly outperformed IFIR across all five subjective image quality categories. Median improvements on the 5‐point scale ranged from 0.6 to 1.0 points. All differences were statistically significant with large effect sizes (|r| = 0.67 to 0.88, corrected p ≤ 0.003), confirming that the observed improvements were clinically meaningful. The greatest advantage was observed in vessel‐to‐background contrast (+1.0 point, |r| = 0.88, corrected p < 0.001), underscoring NSASL's clear superiority for renal arterial vessel visualization. Importantly, these subjective ratings were consistent with the objective CR measurements, providing convergent evidence for the robustness of NSASL's background suppression and arterial enhancement.

TABLE 5.

Subjective comparison of NSASL and IFIR MRA image quality.

Category	IFIR	NSASL	Z	|r|	Raw p‐value	Corrected p‐value
Main renal artery	4.0 (0.7)	4.7 (0.9)	−3.41	0.67	< 0.001	0.003
Branch renal artery	3.3 (0.9)	4.0 (0.6)	−3.60	0.67	< 0.001	0.002
Motion artifacts	3.7 (1.3)	4.3 (1.0)	−3.52	0.67	< 0.001	0.002
Vessel contrast	3.7 (0.3)	4.7 (0.3)	−4.65	0.88	< 0.001	< 0.001
Diagnosis confidence	3.7 (1.3)	4.3 (1.0)	−4.17	0.80	< 0.001	< 0.001

Note: Bolded numbers indicate statistically significant p‐values (p < 0.05).

Abbreviations: IFIR, inflow inversion recovery; NSASL, non‐subtractive arterial spin labeling‐based.

3.4. Preliminary Clinical Feasibility

In six participants undergoing clinical angiography imaging (four with CE‐MRA and two with CTA), qualitative assessments by three blinded radiologists (Tables S7–S10) yielded observations broadly consistent with those in healthy volunteers, although the cohort size precluded meaningful statistical inference. NSASL demonstrated diagnostic confidence scores comparable to CE‐MRA (+ 0.7 points, p = 0.194) and CTA (−0.2 points, p = 0.317). Vessel‐to‐background contrast measurements were similarly comparable, with higher values relative to CE‐MRA (approximately ninefold higher CR, +0.9 points, p = 0.141) and modest difference relative to CTA (around 1.5‐fold higher CR, −0.2 points, p = 0.655), without statistically significant deviations. As illustrated in Figure 5, NSASL generally exhibited clearer background suppression and fewer motion‐related blurring artifacts than IFIR, resulting in comparable or slightly higher reader confidence in some cases. CTA provided the most consistent depiction of distal branches and overall anatomic details, while CE‐MRA occasionally showed reduced branch sharpness or conspicuity. Given the small and clinically heterogeneous sample, these findings should be interpreted as preliminary feasibility observations rather than evidence of diagnostic equivalence or superiority, warranting validation in larger, disease‐focused cohorts.

FIGURE 5.

(a) Renal CE‐MRA coronal MIP of original image generated by Siemens MRI software syngo xA20, (b) CE‐MRA (might have a delayed acquisition timing due to visual observation and manual start‐up), (c) IFIR, and (d) NSASL MIP of cropped images reconstructed by corresponding scanner's software and post‐processed by Matlab 2024b of a 53‐year‐old female patient with hydronephrosis but no renal artery disease before (BMI: 19.7 kg m⁻², RR: 12 breaths per minute, estimated glomerular filtration rate (eGFR): 90.0 mL/min). (e) Renal CTA coronal 3D volume rendering of original image created by post‐processing software uAI Discover Aorta CTA R001 AI‐based automatic vessel segmentation, (f) bone‐removed CTA, (g) IFIR, and (h) NSASL (with a suspected stenosis at right renal artery origin) MIP of cropped images reconstructed by corresponding scanner's software and post‐processed by Matlab 2024b of a 63‐year‐old female patient with high blood pressure and the stenosis at left renal artery origin [Correction added on 9 April 2026, after first online publication: The preceding sentence has been amended.] (BMI: 25.6 kg m⁻², RR: 19 breaths per minute, eGFR: 101.1 mL/min).

4. Discussion

This study introduced and systematically evaluated a novel NC NSASL technique for renal MRA. In healthy volunteers, NSASL consistently outperformed the widely used IFIR method across all assessed categories, including vessel‐to‐background contrast, renal arteries visualization, motion artifacts, and diagnostic confidence. These improvements were not only statistically significant after correction but were also supported by large effect sizes, underscoring their clinical relevance. Exploratory patient data showed that NSASL provided diagnostic information comparable to CE‐MRA and CTA, with distal branch visualization that was similar to CTA and generally better than CE‐MRA, all without the need for contrast agents or radiation; however, these trends require confirmation in larger, disease‐enriched cohorts. The use of different scanner platforms may also have introduced system‐related variability that influenced image quality.

Our findings build on prior reports of IFIR's limitations, including venous contamination, static tissue regrowth, and motion sensitivity [30, 31]. Conventional IFIR typically required a fixed inversion delay (TS of 1200 ms) as a compromise between artery inflow and background suppression, restricting its ability to capture distal branches [32]. Glockner et al. have also shown that these limitations reduce reliability in renal applications [31, 33, 34]. NSASL provided greater flexibility for blood inflow while avoiding the subtraction requirement that hampers conventional ASL‐based approaches. In this way, NSASL effectively merged the strengths of IFIR (background nulling and inflow sensitivity) and pCASL (endogenous blood labeling) while overcoming their respective weaknesses, representing a meaningful technical and clinical advance in NC renal MRA.

Another non‐subtractive ASL method is the Fourier transform‐based velocity‐selective (VS) MRA which suppresses static tissues by selectively labeling blood flowing above a defined velocity threshold and, because of its spatially non‐selective nature, enables large‐FOV imaging independent of inflow direction. This reduces ATT‐related delays and simplifies planning compared with NSASL and IFIR [35]. However, VS‐MRA requires substantially more complex RF pulse design than pCASL and places higher demands on scanner performance, limiting widespread adoption. Its long RF‐gradient pulse trains increase sensitivity to both B0 and B1 inhomogeneity, resulting in phase accrual unrelated to velocity that reduces velocity‐selection accuracy and flip‐angle deviations that diminish labeling efficiency [36]. Additionally, the velocity‐dependent labeling characteristics may lead to incomplete suppression of fast venous blood or unintended suppression of very slow arterial flow in distal branches or severe stenosis, potentially affecting diagnostic reliability [37].

Dong H., et al.'s renal MRA method used multiple global inversion pulses without interleaved pCASL labeling, causing all blood, even outside the imaging slab, to undergo repeated inversions [38]. This led to uniform signal loss of 20%–30% per cycle, and with prolonged inversion times (e.g., 1400 ms), blood signal could drop to below 30%, markedly reducing contrast. Their approach also required both cardiac and respiratory triggering and did not address eddy current‐induced instability in central‐out bSSFP, further limiting robustness. In contrast, NSASL preserves inflowing arterial signals while achieving near‐complete background suppression.

Several technical innovations underpinned NSASL's performance. Unlike IFIR, which relies on a single inversion pulse and is limited by background regrowth at longer ATT, NSASL incorporated multi‐inversion background suppression and interleaved pCASL labeling. This decoupled tissue nulling from ATT, enabling near‐complete suppression of renal parenchyma, venous blood, and other static backgrounds while preserving inflowing arterial signals. In addition, the interleaved small golden remainder angle switching strategy minimized gradient jumps and eddy‐current‐induced bSSFP distortions, enabling central‐out radial acquisition with improved stability. Embedding fat‐suppression modules within FISS readouts, further mitigated the bright fat oscillations that often plague bSSFP. This NSASL labeling module, when combined with the central‐out SoS acquisition trajectory, provided the optimal combination. Collectively, these innovations resolved long‐standing barriers in NC renal MRA, eliminating the need for subtraction and enabling robust free‐breathing imaging.

The magnitude of improvement highlights NSASL's clinical potential. Vessel‐to‐background CR increased nearly fivefold with NSASL, and subjective vessel contrast improved by one full grade, representing very large effect sizes. While CTA and CE‐MRA offered larger abdominal coverage and faster acquisitions, both CTA and NSASL demonstrated clear visualization of third‐ to fourth‐order renal branches, whereas CE‐MRA typically resolved only up to the second order [30, 39, 40]. This branch‐order detail is particularly valuable for detecting subtle vascular abnormalities, but often overlooked with current CE‐MRA due to time‐resolution tradeoff restrictions as well as operation and cooperation requirements. More aggressive GRAPPA acceleration factor in CE‐MRA and CS regulation factor in NSASL might also degrade small vessel visualization. Moreover, NSASL enabled free‐breathing acquisition with shorter scan times than IFIR, reducing dependence on patient cooperation compared with breath‐holding CE‐MRA, and making it potentially advantageous for children and patients with limited compliance [41, 42].

4.1. Limitations

Despite these advantages, several limitations remain. First, SNR was lower in NSASL, primarily due to a higher acceleration factor. Although this was partially compensated by compressed sensing reconstruction, further optimization of trajectory design and gradient calibration may reduce peripheral aliasing [24, 43]. Second, reliance on respiratory triggering may reduce robustness in subjects with irregular or shallow breathing, as the current belt‐based trigger was less sensitive. Integration with navigator‐based retrospective motion weighting or advanced non‐contact sensors could therefore improve reliability [31, 34, 44, 45]. Third, incomplete suppression of very short‐T1 tissues (< 200 ms, e.g., bowel contents) was occasionally observed, where faster readouts, advanced lipid suppression, and better segmentation may ameliorate this [38]. Finally, while central‐out bSSFP sampling preserved vascular contrast better, it inherently applies a k‐space filter that broadens the point spread function, potentially reducing resolution along the partition‐encoding direction; optimized k‐space weighting or advanced iterative reconstructions may help mitigate this effect [46].

Additionally, the patient cohort was small (n = 6) and not enriched for renal vascular disease, precluding evaluation of diagnostic accuracy (e.g., in stenosis). Another limitation was that NSASL and IFIR were evaluated on different MRI platforms. Although NSASL is theoretically compatible with Siemens scanners, we were unable to implement it on the Siemens system due to the absence of a research programming agreement. Conversely, IFIR could not be reproduced on the research system without altering its optimized clinical performance. As a result, cross‐platform comparison was unavoidable. Differences in hardware—such as higher gradient performance, advanced BioMatrix respiratory sensing, and a higher‐channel abdominal coil on the Siemens platform—may have influenced relative performance. These factors likely placed NSASL at a disadvantage, suggesting that even better performance may be achievable once NSASL is deployed on a comparably advanced clinical system.

5. Conclusion

NSASL represents a novel advancement in NC renal MRA. By integrating multi‐inversion suppression, interleaved pCASL labeling, and optimized central‐out SoS bSSFP sampling, NSASL achieved robust background nulling, enhanced arterial conspicuity, improved motion tolerance, and greater diagnostic confidence compared with IFIR. Effect size analyses confirmed that these improvements were not only statistically significant but also clinically relevant. While further validation in a larger, pathologically diverse patient cohort is still needed, NSASL showed strong potential as a safe, non‐invasive alternative to CE‐MRA and CTA, particularly for patients with contraindications of gadolinium or ionizing radiation.

Funding

This study was supported by the National Key R&D Program of China: 2022YFC240890, Ningbo “S&T Innovation 2035” Major programs: 2023Z182 and 2025Z20.

Supporting information

Figure S1: Reconstructed NSASL image MIP without CS, with CS using λ of 0.5%, 1.0%, 1.5%, 2.0% of the maximum normalized pixel value from left to right of two healthy volunteers: a 25‐year‐old man (BMI: 21.6 kg m⁻², RR: 12 breaths per minute) (upper) and a 29‐year‐old woman (BMI: 22.7 kg m⁻², RR: 28 breaths per minute) (bottom).

Figure S2: MIP images of NSASL and IFIR with (a) the same trigger number of 72 and (b) the same 0.75 partial Fourier of a healthy 27‐year‐old woman (BMI: 25.1 kg m⁻², RR: 17 breaths per minute).

Figure S3: MIP images of different preparations, acquisition trajectories and orders of a healthy 27‐year‐old woman (BMI: 25.1 kg m⁻², RR: 17 breaths per minute).

Figure S4: The MIP images of renal arteries obtained with IFIR and NSASL MRA in transversal, sagittal, and coronal planes collected with different TS values (800–1600 ms) of a healthy 27‐year‐old woman (BMI: 25.1 kg m⁻², RR: 18 breaths per minute). The NSASL had stable and superior background suppression and artery visualization, whereas IFIR suffered blood signal decrease with low TS and background regrowth with high TS.

TABLE S1: Five‐point Image quality grading scale.

TABLE S2: Demographic characteristics of volunteers underwent CE‐MRA or CTA (n = 6).

TABLE S3: Scan values of NSASL and IFIR with the same trigger number of 72.

TABLE S4: Scan values of NSASL and IFIR with the same 0.75 partial Fourier.

TABLE S5: Cartesian and SoS acquisition parameters.

TABLE S6: Subjective comparison among different module combinations.

TABLE S7: Objective comparison of NSASL, IFIR and CE‐MRA image quality of 4 participants with diseases.

TABLE S8: Objective comparison of NSASL, IFIR and CTA image quality of 2 participants with diseases.

TABLE S9: Subjective comparison of NSASL, IFIR, and CE‐MRA image quality of 4 participants with diseases.

TABLE S10: Subjective comparison of NSASL, IFIR, and CTA image quality of 2 participants with diseases.