Navigator-gated free-breathing joint T1-T2 mapping of the kidney

Purpose To develop and evaluate a free -breathing 2D radial joint T ₁-T₂ mapping technique for the kidneys at 3T, and to assess the impact of navigator gating parameters on mapping precision and accuracy.

The PARMANav sequence (PArametric Radial M Apping with Navigator gating) was implemented for renal imaging, using 25 single-shot radial gradient echo acquisitions with five repeated magnetization preparations and lung -liver navigator gating to avoid through -plane motion. Virtual compressed coil and c ompressed sensing with spatial and contrast regularization was used for image reconstruction, followed by a model-based registration. An acquisition-specific joint T ₁-T₂ dictionary was generated using extended phase graph simulations. T1 and T2 accuracies were quantified in a phantom study versus gold standard spin-echo-based sequences. T he influence of the navigator acceptance window width (NAWW) and navigator rejection on T 1 and T 2 precision were established in 10 healthy volunteers and were compared to routine T1 and T2 mapping. Three patients were scanned to demonstrate clinical feasibility. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.

In the phantom, PARMANav T1 and T2 values showed high accuracy with the gold standard T1 and T2 values and were insensitive to rejected navigators (< 5% variation for T1 and T2). As expected from previous studies, in-vivo renal PARMANav T1 and T2 values were higher than routine values but showed lower variability, both per subject and between subjects : in the cortex PARMANav T1=1601±48ms/T2=90.8±5.0ms vs routine T1=1307±108ms/T2=73.3±8.0ms, while in the medulla PARMANav T 1=2044±82ms/T2 =90.3±5.4ms and routine T1=1560±122ms/T2=67.6±5.8ms. No T1 or T2 trend was observed for the different NAWW. Feasibility was demonstrated in patients, where high-quality maps were obtained.

PARMANav allows for precise and accurate joint T 1-T2 mapping of the kidneys without requiring breath holding. Through-plane motion artifacts were avoided with a navigator, which did not impact the accuracy or precision of the resulting maps.

In kidney disease, the T1 and T 2 relaxation times reflect underlying changes in tissue composition,1 while a correlation between T1 cortico-medullary differentiation (CMD) and renal function has been demonstrated in several renal diseases.2,3 T1 CMD was proven to be related to the degree of fibrosis, while elevated cortex T1 values is strongly associated with poor renal outcome in patient with chronic kidney disease and with allograph kidneys,4 and oedema and T2 increases with inflammation and edema. 5 Renal T 1 mapping is typically performed with modified Look-Locker inversion recovery (MOLLI),5,6 which is known to underestimate the T 1 value.7 Renal T 2 mapping is less common and is usually performed with a turbo spin -echo sequence, which typically overestimates the T2 value. Instead of mapping a single relaxation time per acquisition, o ver the past decade, multiparametric mapping techniques have enabled the simultaneous measurement of multiple relaxation times in a single acquisition. These approaches have gained interest thanks to their ability to provide more comprehensive insights into renal structure and function,5 while known pulse sequence imperfections (e.g., slice profile, inversion efficiency) can be incorporated in the fitting model to improve accuracy. Although such techniques have been successfully commercialized for neuroimaging with MR fingerprinting,8 and a large number of variations has been studied for cardiac MRI ,9–12 its application to renal imaging is thus far limited to the research setting, partly because they require breath holding, which is not always feasible in kidney disease patients. The development of free -breathing techniques would represent a n . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint impactful step forward towards the integration of T 1 and T 2 mapping in clinical practice to assess kidney structure in these patients. Recent studies have demonstrated applications for breath-held joint T1-T2* mapping13,T1-T2 mapping14, and free-breathing T1-T2* mapping15 of the kidney at 3T. Most of the abovementioned techniques require breath holding, which is not always feasible in patients . Free-breathing archieved with respiratory gating, 15 relies on the assumption of a constant respiratory cycle duration and used a two parameters analytical fit, that doesn’t allow to model precisely the magnetization evolution. Conversely, many recent multiparametric techniques claim to allow for free breathing through the use of state-of-the-art in-plane retrospective motion correction (i.e., registration) between the source images , but do not account for through-plane motion: motion in the direction perpendicular to the image plane and thus invisible to a motion correction algorithm . This omission could be significant, potentially lead ing to perceived increased values in healthy tissues or normal values in a lesion. The reliance on registration, especially non -rigid registration, could also induce additional errors that propagate in final maps. To enable a free- breathing acquisition that avoids such discrepancies between its source images, a lung-liver navigator can be used to track the diaphragm position ,9 albeit at the cost of acquisition efficiency. The accuracy of such a technique was previously assessed in a cardiac numerical phantom and reported no dependencies on the number of navigator rejections.16 In the current study, we aimed to demonstrate that an adaptation of the free -breathing navigator-gated multi parametric mapping technique PARMANav (PArametric Radial MApping with Navigator gating)16 can be used to obtain accurate and precise parametric maps of the kidney at 3T while avoiding through-plane motion.

Pulse sequence design We adapted PARMANav16 for renal imaging by implementing a free-breathing acquisition of 25 magnetization -prepared single -shot 2D gradient -recalled echo (GRE) images , e ach acquired with a continuous golden angle trajectory (Figure 1) . To enable magnetization recovery, the blocks (i.e., a magnetization preparation and a single-shot image acquisition) were repeated every 1s. To achieve joint T₁-T₂ sensitivity, five different contrast preparations were repeated in series : one adiabatic inversion pulse ( a 5.12 ms hyperbolic tangent), no preparation, and three different T₂-preparation modules. This set was repeated five times to ensure contrast diversity throughout the different number of skipped navigators (Figure 1 A). . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint Figure 1. Overview of the free -breathing radial 2D joint T 1-T2 mapping technique PARMANav. A) The p ulse sequence diagram shows the first of five repeated blocks that consist of five differently prepared images, and the simulated magnetization of a healthy renal cortex and medulla. B) Illustration of the placement of the navigator (green) and image slice (yellow) as well as a trace of the navigator with the acceptance window as two parallel green lines. C) Example first 5 images in a healthy volunteer kidney, reconstructed using compressed . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint sensing (CS). D) The dictionary is created through EPG simulations of the magnetization across the 25 images. E) Model-based motion correction between source and synthetic images obtained with the dictionary and the first unregistered maps. F) The final maps are obtained by computing the pixel-wise dot-product between the registered images and the dictionary 𝑑⃑. Respiratory motion was tracked using a lung-liver navigator acquired before each preparation, with slice tracking enabled. To ensure that the timing of all magnetization changes was known, both preparation and readout modules were skipped when navigator rejection occurred. All data were acquired on a 3T clinical scanner (Magnetom Prisma or PrismaFit, Siemens Healthineers, Forchheim, Germany) with nominal matrix size 192x192 (resulting in a 384x384 matrix through oversampled radial gridding), 45 continuous golden-angle (68°) radial lines per image (corresponding to 15% radial Nyquist sampling), pixel size=(1.56mm)2, slice thickness 8mm, flip angle 12°, bandwidth 789Hz/pixel, repetition time TR 3.49ms, echo time TE 1.56ms, acquisition window duration 151ms, inversion time TI 68ms (for the image directly after the inversion), and T2-preparartion modules echo times 23/45/70 ms. A fixed 5s delay at the start of the pulse sequence allows for complete magnetization relaxation between successive acquisitions. For all acquisitions we used a 34-channel chest-spine coil array. Image and map reconstruction Individual RF coil elements were combined using region-optimized virtual (ROVir) coils17,18 to minimize radial streaking artifacts coming from the arms and the abdominal fat. Briefly, the 192x192-pixels central part of the acquired image was automatically selected as the region of interest, and the periphery of the image was designated as the unwanted signal region. A generalized eigenvalue decomposition was used to identify virtual coils that maximize the signal-to-interference ratio (SIR). The smallest set of virtual coils capturing ≥90% of the total signal energy was retained. From these undersampled k -spaces, images were reconstructed using compressed sensing with total variation regularization in the spatial dimension and local -low-rank regularization along the contrast dimension:19 𝐱̂ = 𝐚𝐫𝐠 𝐦𝐢𝐧𝐱 ‖𝐅𝐂𝐱 − 𝐲‖𝟐 𝟐 + 𝝀𝐒‖𝛁𝐬𝐱‖𝟏 + 𝝀𝐂 ∑ ‖𝐋𝒊𝐱‖𝒊>𝟏 ∗ (1), where 𝐱̂ is the reconstructed image, 𝐅 is the nonuniform fast Fourier operator, 𝐂 is the coil sensitivity, 𝐲 refers to the acquired k-space, 𝛁𝑺 is the first-order difference operator along the spatial dimension, 𝐋𝒊 is the operator that extracts the i-th spatial patch of size 6x6x6 pixels . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint from 𝐱 and forms a Casorati matrix with the contrast dimension, ‖∙‖∗ is the nuclear norm, and 𝜆𝑆 and 𝜆𝐶 are the corresponding regularization weights along the spatial and contrast dimensions, respectively. Regularization parameters, 𝜆𝑆 = 0.01 and 𝜆𝐶 = 0.06 were empirically selected for optimal trade-off between undersampling artifact removal and image blurring. Considering individual navigator rejections, a n acquisition-specific signal dictionary was generated via extended phase graph simulations in MATLAB ( version R2023b, The Mathworks, Natick, Massachusetts, USA) across a wide range of T₁ values from 0 ms to 5000 ms in 10 ms increments, and from 5020 ms to 6000 ms in 20 ms increments. and T₂ values from 0 ms to 450 ms with a 5 ms increment and from 460 ms to 700 ms with a 10 ms increment, incorporating slice profile effects ( discretized in 50 isochromats) and inversion inefficiency correction.20 The precise acquisition timing was extracted from raw data to reflect navigator acceptance. The magnetization was simulated at the center of each echo, and the signal averaged over each radial single-shot echo train, resulting in a 25×60 396 (T2>T1 cases being excluded) matrix of simulated complex signals that formed the dictionary. A previously described model-based non-rigid image registration16 was applied to account for residual in-plane motion that resulted from residual differences in the respiratory phase while accounting for the strong contrast variations in these images (Figure 1 E). Here, a first set off maps was generated via pixel -wise dictionary matching without motion correction. Synthetic images with matched contrast and averaged motion were then created and used as references for non -rigid optical -flow-based registration. 21 Final maps were obtained by repeating dictionary matching on the motion-corrected images. Phantom study The “ISMRM/NIST” phantom22 (Premium System 130, CaliberMRI, Boulder, USA) was scanned to evaluate the accuracy of PARMANav readout echo trains were separated by intervals of 1s, which were sporadically extended to 2s or 3s such that navigator rejections could be emulated. Clinical routine T 1 and T2 maps (pixel size=(1.4-1.9mm)2, slice thickness 8mm) were acquired using 5(3)5 MOLLI,6 and T2-prepared (T2-prep) bSSFP T2 mapping,23,24 respectively. It was compared against the values obtained with gold-standard inversion - recovery spin -echo (for T 1 relaxation) and spin -echo (for T 2 relaxation) techniques. The different sequences parameters are reported in Table 1. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint Table 1. Parameters of the sequences used for T1 and T2 mapping. The same parameters were used in the phantom and in-vivo scans. PARMANav MOLLI T2-prep bSSFP Resolution 1.56 x 1.56 mm2 1.4 x 1.4 mm2 1.9 x 1.9 mm2 Matrix size 192 x 192 256 x 141 192 x 117 Slice thickness 8 mm Flip angle 12 ° 12 ° 60 ° Bandwidth 789 Hz/pixel 977 Hz/pixel 930 Hz/pixel TR 3.49 ms TE 1.56 ms 1.18 ms 1.04 Preparation pulse Inversion, 3 T2-prep Inversion 3 T2-prep Trajectory Golden-angle radial Cartesian Acceleration Radial 6.7x GRAPPA 2x Free-breathing Yes, with a navigator Breath-held (13 s) Breath-held (9 s) Healthy volunteer study Ethics approval was obtained from the Ethics Committee of the Canton of Vaud (CER-VD) of Switzerland under reference numbers 2021-00697 and 2022-00934. All participants provided written informed consent to participate prior to enrollment. PARMANav maps of N=10 healthy volunteers (31.8±9.1 y, 4F) were acquired with four different navigator acceptance window widths (NAWWs of ±4mm, ±8mm, ±16mmm, and ±32mm) of the lung-liver navigator in a randomized order to study the impact of through -plane motion. Volunteers were instructed to breath normally. Like in the phantom study, c linical routine breath-held T1 and T2 maps were acquired using MOLLI and T2-prepared bSSFP T2 mapping, respectively. The T1 and T2 value of the visible cortex and medulla as well as the CMD ratio (as the ratio T1cortex/T1medulla) were determined for each map in each volunteer by manually segmenting regions of interest (ROIs) in both kidneys on the T 1 map. Values from the two kidneys were averaged. The coefficient of variation (CoV) was calculated as the regional standard deviation divided by the mean relaxation time. The acquisition time was recorded for the four NAWWs. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint Patient study To preliminarily evaluate its clinical feasibility, PARMANav was acquired in three patients (65.7±15.0,1F) as part of an ongoing study on heart failure.25 Two patients had chronic kidney disease (CKD), one with heart failure with preserved ejection fraction (HFpEF) and one with heart failure with reduced ejection fraction, while one patient had HFpEF but no CKD. In each patient, one map was acquired with a NAWW of ±8mm. No routine mapping techniques were acquired due to the constraints of the total MRI protocol duration. Anatomical reference images (T2-weighted HASTE) were acquired at the same location for two of the three patients, and at a slightly different orientation for the last patient. Statistics Agreement between the measured and gold-standard values in the phantom was evaluated using the slope and coefficient of determination (R²) from linear regression analysis for both PARMANav with and without skipped navigator, and with the routine techniques. Bland- Altman analysis was performed to quantify biases and limits of agreement, and a paired t- test was conducted to evaluate statistical significance. In the healthy volunteers, the T1 and T2 values of the cortex and medulla were extracted for the four NAWWs and the two routines methods. A Shapiro-Wilk test was performed to assess normality. Their differences, as well as the time of acquisition differences, were tested with a repeated-measures analysis of variance (RM-ANOVA) with post-hoc Tukey analysis. The regional T1-T2 values and CoV means and standard deviations (SDs) across the healthy volunteers were reported. Bland-Altman analysis was performed to quantify biases and limits of agreement.

Phantom study In the phantom, PARMANav showed high agreement with the gold standard, both with and without rejected navigators (Nskipped =23) (Figure 2): the slope of the correlation was closer to identity than the routine technique for T1 (1.05 and 1.00 for without and with skipped heartbeat, respectively vs 0.76 for MOLLI), while there was a larger difference for T2 (1.10 and 1.21 vs 0.60). R² was above 0.99 for both the PARMANav scans and MOLLI, and slightly different for the T2 mapping methods (>0.99 for the two PARMANav vs 0.94 for T2- prep bSSFP). As with previous studies,16 PARMANav with and without skipped navigators did not significantly differ for T2 (P>0.1). A significant difference was reported for T1 (P=0.005), with a small average relative difference (6.6%). . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint Figure 2. Agreement of PARMANav T 1 and T 2 maps in the ISMRM-NIST phantom compared to spin-echo reference values in case of 23 skipped navigator. A) T1 map of the phantom obtained with PARMANav. B) T2 map of the reference phantom obtained with PARMANav. C) Linear regression of PARMANav T1 values with and without rejected navigator and the clinical routine 5(3) 5 MOLLI versus gold -standard IR-SE. D) Linear regressions of PARMANav T 2 values with and without rejected navigator and the clinical routine T 2-prep bSSFP in the phantom in the clinically relevant range versus gold -standard SE. E) T1 Bland- Altman plot of PARMANav with and without rejected navigator and MOLLI versus gold - standard IR-SE. The bias and confidence bounds are reported in the legend . PARMANav presented a larger bias when adding a large number of skips (N=23). F) Same as E) for T 2 with T2-prep bSSFP as the clinical routine and SE as the gold-standard. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint Precision in Healthy Volunteers In all 10 healthy volunteers, visually sharp and precise maps of the kidney were obtained with PARMANav (Figure 3). PARMANav T1-T2 values and CMD were significantly different from routine techniques, which are known to underestimate T1 and overestimate T2,26 but there were no significant differences as a function of the differ ent NAWWs (Figure 4). The cortex and medulla T1 values presented a larger spread for NAWW=±4 mm than for NAWW=±8 mm with more outliers (SD4mm = 117 ms and SD8mm = 48 ms for the cortex). NAWW=±4 mm also resulted in a significantly longer acquisition time (p<0.02 for NAWW=±4 mm versus all the other NAWWs)(Figure 5). Given these results and the need for balance between low navigator acceptance (at small NAWW) and tolerated through-plane motion (at large NAWW s), NAWW=±8 mm mapping was chosen as best compromise for further analysis. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint Figure 3. PARMAN av T1 and T 2 maps for the different NAWWs compared to the routine techniques. The NAWW did not significantly change the relaxation times in the resulting PARMANav T1 or T 2 maps of the cortex, with the exception of the poles (red arrowheads), where partial volume effect lower ed the T 1 relaxation time. However, the shape and the T1 values of the medulla varied with the NAWW (orange arrowhead), likely due to through-plane motion. The PARMANav T1 and T2 values were consistently higher than those obtained with the routine techniques. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint Figure 4. Impact of the navigator acceptance window width (NAWW) . in 10 healthy volunteers. A-C) T1 values the cortex and in the medulla and the corresponding CMD ratio for the four NAWWs compared to MOLLI, which is known to underestimate T1 value, especially at these higher values . D, E) T2 values in the cortex and in the medulla for the four NAWW compared to T2-prepared bSSFP. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint Figure 5. PARMANav a cquisition time for the different NAWWs. Acquisition time was longer for smaller NAWW as expect ed. For NAWW=±16 and 32 mm, the time of acquisition was almost identical, due to no rejection for both NAWWs . The difference was significant between all the time of acquisition (p<0.02) except between NAWW=±16 and 32 mm (p=0.79) PARMANav T1-T2 values were higher than routine techniques (Table 2). Smaller inter-subject SD was reported for all relaxation times. Table 2. Cortex and medulla T1 and T2 mean and standard deviation in ms across the 10 healthy volunteers, as well as the coefficient of variation (CoV). PARMANav NAWW=±8 mm Routine Mean ± SD CoV Mean ± SD CoV Cortex T1 1601 ± 48 ms 4.1 ± 0.7 % 1307 ± 108 ms 5.3 ± 2.0 % Medulla T1 2044 ± 83 ms 4.0 ± 0.7 % 1434 ± 171 ms 5.4 ± 1.8 % Cortex T2 90.8 ± 5.0 ms 6.3 ± 2.4 % 73.3 ± 8.0 ms 4.3 ± 1.3 % Medulla T2 90.3 ± 5.4 ms 7.4 ± 3.4 % 67.6 ± 5.8 ms 7.2 ± 5.4 % The Bland Altman analysis (Figure 6) demonstrated a bias against the reference techniques, which is consistent with the phantom results and segmental averages. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint Figure 6. Bland -Altman analyses of the agreement between PARMAN av with NAWW=±8mm and the reference techniques. A) Cortex T1 values. B) Cortex T2 values. C) Medulla T1 values. D) Medulla T2 values. Feasibility in Patients PARMANav was successfully acquired in the three patients and resulted in sharp and artefact-free maps (Figure 7). A cyst, also visible on the localizer image could be clearly observed on the maps for Patient 1 – the free liquid interior results in very high T1 and T2 relaxation times. The individual structures in the kidney are more difficult to differentiate in all three patients, while they had a CMD ratio close to one (CMD ratio = 0.80,0.87 and 0.89 for patient 1,2 and 3, respectively) as expected.2,3 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint Figure 7. Parametric PARMANav maps obtained in patients. Patient 1 had chronic kidney disease (CKD) and heart failure with preserved ejection fraction (HFpEF) as well as a cyst (red arrowhead), Patients 2 had chronic kidney disease and heart failure with reduced ejection fraction, and Patient 3 had HFpEF without CKD. A T2 HASTE anatomical image is provided for anatomical reference.

In this work, a free -breathing 2D technique was characterized for joint T 1–T2 mapping of the kidneys at 3T. Previous numerical simulations have shown that the estimated T1 and T2 values were not impacted by the number of rejected navigators16. The mapping phantom accuracy obtained in this study further support this finding. PARMANav resulted in precise free-breathing joint T 1-T2 maps of the kidney in the healthy volunteers. It was successfully acquired in all healthy volunteers with all tested NAWWs, which resulted in similar mapped relaxation times. In patients, free -breathing joint T1-T2 maps . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint resulted in artefact -free maps of high quality with the added advantage that the T1 and T2 maps are intrinsically registered. The Bland -Altman study in the healthy volunteers showed a bias compared to routine techniques, which was relatively high for the medulla T 1 values, but agreed well with the phantom study. T1 and T2 values were systematically higher for PARMANav, both in the cortex and medulla, with large limit of agreement. The higher T1 SD for NAWW=± 4 mm might indicate that a higher number of navigator rejections impacts the T1 precision. To mitigate this effect, a variable flip angle could be introduced for the acquisition, which could also be used to calculate a B1 map.11 Although the impact of the number and timing of rejected navigators (e.g., immediately following the inversion pulse) assessed in a mapping phantom would be valuable, the current experimental setup did not permit such precise control. Future studies using a motion phantom may enable these investigations and facilitate evaluation of accuracy in the presence of motion. No consistent relationship was observed between NAWW and T1-T2 precision or accuracy, although this is likely due to avoiding clear artefactual borders during the segmentation process (as illustrated in Figure 3). This suggests that NAWW selection may be guided primarily by the trade-off between scan duration and through -plane motion tolerance. However, the shape of the cortex and medulla varied with the larger NAWW s, likely due to partial volume effect s or

in the registration algorithm under conditions of increased motion. The increased number of outliers observed with NAWW=±4mm motivated the selection of an ±8 mm NAWW for subsequent experiments. Additional parameters, such as the flip angle and the number of radial lines per source image could be experimentally optimized and potentially reduced in future studies. Several breath-held magnetic resonance fingerprinting (MRF) techniques have been proposed for renal mapping, all relying on Bloch equation simulations. Chen et al.⁸ developed an abdominal fingerprinting technique for simultaneous T ₁ and T₂ mapping with B ₁ correction. Compared to this method, which used spiral sampling and Bloch-based dictionary simulation, PARMANav yielded higher T ₁ and T₂ values in both the cortex and medulla. More recently, Hermann et al.⁹ introduced a breath -held MRF sequence for T ₁ and T₂ mapping across four slices. Their reported T₁ values, while still slightly lower, were more closely aligned with those obtained using PARMANav (T1 = 1456 ms and 1921 ms in cortex and medulla, respectively, vs. 1601 ms and 2044 ms with PARMANav). MacAsk ill et al.¹⁰ also presented a breath -held . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint kidney MRF technique for T ₁ and T ₂ mapping with B ₁ correction, again reporting shorter relaxation times than those measured using PARMANav. A free -breathing method was proposed by Ding et al.15 for T1 and T2* mapping, using respiratory triggering via a respiratory belt. The mapping was based on an analytical equation, which does not allow to model the imperfection of the acquisition (e.g., RF profile, inversion efficiency) and assumes a constant respiratory cycle duration. The images acquired in patients were sharp and artefact-free. These results are encouraging and suggest that combined T1-T2 free breathing mapping with NAWW is possible in CKD patients, whether or not will they suffer from associated heart fail ure. As a next step, PARMANav should be compared to standard techniques. Future studies should also include a larger number of CKD patients, with different associated comorbidities known to lead to disturbances in breathing such as underlying lung disease or morbid obesity. This study has several limitations. As mentioned, the number of included patients was small. Besides, dictionary-based multiparametric mapping is limited by discretization and computational demands: coarse grids introduce small errors, while finer grids increase generation and matching time. Other limitations of the proposed technique include challenges in sequence optimizations, as navigator rejections may have a greater impact on contrast variability than the parameter being optimized. Due to the complexity of kidney anatomy, only relatively small ROIs can typically be manually segmented. Although a second ROI was drawn on the second kidney to mitigate those effect, automated segmentation of the entire cortex and medulla could provide more representative and robust measurements .27 Finally, further developments might include semi-automated segmentation, an extension to T2* mapping28 to more sensitively assess oxygenation, and diffusion modules to further characterize fibrosis.

We demonstrated that the proposed navigator -gated 2D radial GRE sequence PARMANav enables accurate and precise simultaneous T₁-T₂ mapping of the kidneys during free breathing at 3T. The navigator acceptance window width (NAWW) had minimal impact on the accuracy, although very narrow or wide windows introduced more outliers, which led to map degradation, potentialy due to residual motion and registration errors. Based on these findings, an NAWW of ±8 mm was selected as a trade-off between scan efficiency and motion robustness. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint

This study was funded by the Swiss National Science Foundation (SNSF) under grant number CRSII5_202276.

1. Dekkers IA, de Boer A, Sharma K, et al. Consensus-based technical recommendations for clinical translation of renal T1 and T2 mapping MRI. MAGMA. 2020;33(1):163-176. doi:10.1007/s10334-019-00797-5 2. Marotti M, Hricak H, Terrier F, McAninch JW, Thuroff JW. MR in renal disease: importance of cortical-medullary distinction. Magn Reson Med. 1987;5(2):160-172. doi:10.1002/mrm.1910050207 3. Hricak H, Terrier F, Demas BE. Renal allografts: evaluation by MR imaging. Radiology. 1986;159(2):435-441. doi:10.1148/radiology.159.2.3515420 4. Huber A, Aslam I, Crowe L, et al. T1 mapping magnetic resonance imaging predicts decline of kidney function. Clin Kidney J. 2025;18(3):sfaf032. doi:10.1093/ckj/sfaf032 5. Wolf M, de Boer A, Sharma K, et al. Magnetic resonance imaging T1- and T2-mapping to assess renal structure and function: a systematic review and statement paper. Nephrol Dial Transplant. 2018;33(suppl_2):ii41-ii50. doi:10.1093/ndt/gfy198 6. Messroghli DR, Radjenovic A, Kozerke S, Higgins DM, Sivananthan MU, Ridgway JP. Modified Look-Locker inversion recovery (MOLLI) for high-resolution T1 mapping of the heart. Magnetic Resonance in Medicine. 2004;52:141-146. doi:10.1002/mrm.20110 7. Kellman P, Hansen MS. T1-mapping in the heart: accuracy and precision. Journal of Cardiovascular Magnetic Resonance. 2014;16(1):2. doi:10.1186/1532-429X-16-2 8. Ma D, Gulani V, Seiberlich N, et al. Magnetic resonance fingerprinting. Nature. 2013;495(7440):187-192. doi:10.1038/nature11971 9. Hamilton JI, Jiang Y, Chen Y, et al. MR fingerprinting for rapid quantification of myocardial T1 , T2 , and proton spin density. Magn Reson Med. 2017;77(4):1446-1458. doi:10.1002/mrm.26216 10. Chow K, Hayes G, Flewitt JA, et al. Improved accuracy and precision with three-parameter simultaneous myocardial T1 and T2 mapping using multiparametric SASHA. Magnetic Resonance in Medicine. 2022;87(6):2775-2791. doi:10.1002/mrm.29170 11. Henningsson M. Cartesian dictionary-based native T1 and T2 mapping of the myocardium. Magnetic Resonance in Medicine. 2022;87(5):2347-2362. doi:10.1002/mrm.29143 12. Xanthis CG, Bidhult S, Greiser A, et al. Simulation-based quantification of native T1 and T2 of the myocardium using a modified MOLLI scheme and the importance of Magnetization Transfer. Magnetic Resonance Imaging. 2018;48:96-106. doi:10.1016/j.mri.2017.12.020 13. Hermann I, Chacon-Caldera J, Brumer I, et al. Magnetic resonance fingerprinting for simultaneous renal T1 and T2* mapping in a single breath-hold. Magn Reson Med. 2020;83(6):1940-1948. doi:10.1002/mrm.28160 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint 14. MacAskill CJ, Markley M, Farr S, et al. Rapid B1-Insensitive MR Fingerprinting for Quantitative Kidney Imaging. Radiology. 2021;300(2):380-387. doi:10.1148/radiol.2021202302 15. Ding Y, Mason RP, McColl RW, et al. Simultaneous measurement of tissue oxygen level- dependent (TOLD) and blood oxygenation level-dependent (BOLD) effects in abdominal tissue oxygenation level studies. Journal of Magnetic Resonance Imaging. 2013;38(5):1230-1236. doi:10.1002/jmri.24006 16. Calarnou P, Ogier AC, Roy CW, et al. Navigator-gated free-breathing 2D radial cardiac joint T1– T2 mapping. Magnetic Resonance in Medicine. n/a(n/a). doi:10.1002/mrm.30588 17. Kim D, Cauley SF, Nayak KS, Leahy RM, Haldar JP. Region-optimized virtual (ROVir) coils: Localization and/or suppression of spatial regions using sensor-domain beamforming. Magn Reson Med. 2021;86(1):197-212. doi:10.1002/mrm.28706 18. Tasdelen B, Lee NG, Cui SX, Nayak KS. Improved abdominal T1 weighted imaging at 0.55T. Magnetic Resonance in Medicine. 2024;92(6):2580-2587. doi:10.1002/mrm.30224 19. Di Sopra L, Roy CW, Bastiaansen JAM, et al. Fully Self-Gated Cardiac and Respiratory Motion- Resolved Isotropic 5D T1 Mapping of the Heart: Preliminary Results. In: Proc Int Soc Magn Reson Med 27. Montréal QC, Canada; 2019. 20. Hamilton JI, Jiang Y, Ma D, et al. Investigating and Reducing the Effects of Confounding Factors for Robust T1 and T2 Mapping with Cardiac MR Fingerprinting. Magn Reson Imaging. 2018;53:40-51. doi:10.1016/j.mri.2018.06.018 21. Horn BKP, Schunck BG. Determining optical flow. Artificial Intelligence. 1981;17(1):185-203. doi:10.1016/0004-3702(81)90024-2 22. Stupic KF, Ainslie M, Boss MA, et al. A standard system phantom for magnetic resonance imaging. Magnetic Resonance in Medicine. 2021;86(3):1194-1211. doi:10.1002/mrm.28779 23. Giri S, Chung YC, Merchant A, et al. T2 quantification for improved detection of myocardial edema. J Cardiovasc Magn Reson. 2009;11:56. doi:1532-429X-11-56 [pii] 10.1186/1532-429X- 11-56 24. van Heeswijk RB, Feliciano H, Bongard C, et al. Free-Breathing 3 T Magnetic Resonance T2- Mapping of the Heart. JACC: Cardiovascular Imaging. 2012;5(12):1231-1239. doi:10.1016/j.jcmg.2012.06.010 25. Meyer P, Rocca A, Banus J, et al. Characterizing subtypes of heart failure with preserved ejection fraction: the HeartMagic prospective observational study. April 2025:2025.04.10.25325567. doi:10.1101/2025.04.10.25325567 26. Messroghli DR, Moon JC, Ferreira VM, et al. Clinical recommendations for cardiovascular magnetic resonance mapping of T1, T2, T2* and extracellular volume: A consensus statement by the Society for Cardiovascular Magnetic Resonance (SCMR) endorsed by the European Association for Cardiovascular Imaging (EACVI). J Cardiovasc Magn Reson. 2017;19(1):75. doi:10.1186/s12968-017-0389-8 27. Aslam I, Aamir F, Kassai M, et al. Validation of automatically measured T1 map cortico- medullary difference (ΔT1) for eGFR and fibrosis assessment in allograft kidneys. PLOS ONE. 2023;18(2):e0277277. doi:10.1371/journal.pone.0277277 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint 28. Li LP, Milani B, Pruijm M, et al. Renal BOLD MRI in patients with chronic kidney disease: comparison of the semi-automated twelve layer concentric objects (TLCO) and manual ROI methods. Magn Reson Mater Phy. 2020;33(1):113-120. doi:10.1007/s10334-019-00808-5 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint Figure Captions Figure 1. Overview of the free-breathing radial 2D joint T1-T2 mapping technique PARMANav. A) The pulse sequence diagram shows the first of five repeated blocks that consist of five differently prepared images, and the simulated magnetization of a healthy renal cortex and medulla. B) Illustration of the placement of the navigator (green) and image slice (yellow) as well as a trace of the navigator with the acceptance window as two parallel green lines. C) Example first 5 images in a healthy volunteer kidney, reconstructed using compressed sensing (CS). D) The dictionary is created through EPG simulations of the magnetization across the 25 images. E) Model-based motion correction between source and synthetic images obtained with the dictionary and the first unregistered maps. F) The final maps are obtained by computing the pixel-wise dot-product between the registered images and the dictionary d. Figure 2. Agreement of PARMANav T1 and T2 maps in the ISMRM-NIST phantom compared to spin-echo reference values in case of 23 skipped navigator. A) T1 map of the phantom obtained with PARMANav. B) T2 map of the reference phantom obtained with PARMANav. C) Linear regression of PARMANav T1 values with and without rejected navigator and the clinical routine 5(3)5 MOLLI versus gold -standard IR-SE. D) Linear regressions of PARMANav T 2 values with and without rejected navigator and the clinical routine T 2-prep bSSFP in the phantom in the clinically relevant range versus gold -standard SE. E) T1 Bland-Altman plot of PARMANav with and without rejected navigator and MOLLI versus gold-standard IR-SE. The bias and confidence bounds are reported in the legend . PARMANav presented a larger bias when adding a large number of skips (N=23). F) Same as E) for T2 with T2-prep bSSFP as the clinical routine and SE as the gold-standard. Figure 3. PARMANav T1 and T2 maps for the different NAWWs compared to the routine techniques. The NAWW did not significantly change the relaxation times in the resulting PARMANav T1 or T2 maps of the cortex, with the exception of the poles (red arrowheads), where partial volume effect lowered the T1 relaxation time. However, the shape and the T1 values of the medulla varie d with the NAWW (orange arrowhead), likely due to through - plane motion. The PARMAN av T1 and T 2 values were consistently higher than those obtained with the routine techniques. Figure 4. Impact of the navigator acceptance window width (NAWW). in 10 healthy volunteers. A-C) T1 values the cortex and in the medulla and the corresponding CMD ratio for the four NAWWs compared to MOLLI, which is known to underestimate T 1 value, especially at these higher values. D, E) T2 values in the cortex and in the medulla for the four NAWW compared to T2-prepared bSSFP. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint Figure 5. PARMANav acquisition time for the different NAWWs. Acquisition time was longer for smaller NAWW as expect ed. For NAWW=±16 and 32 mm, the time of acquisition was almost identical, due to no rejection for both NAWWs. The difference was significant between all the time of acquisition (p<0.02) except between NAWW=±16 and 32 mm (p=0.79) Figure 6. Bland-Altman analyses of the agreement between PARMANav with NAWW=±8mm and the reference techniques. A) Cortex T1 values. B) Cortex T2 values. C) Medulla T1 values. D) Medulla T2 values. Figure 7. Parametric PARMANav maps obtained in patients . Patient 1 had chronic kidney disease (CKD) and heart failure with preserved ejection fraction (HFpEF) as well as a cyst (red arrowhead), Patients 2 had chronic kidney disease and heart failure with reduced ejection fraction, and Patient 3 had HFpEF without CKD. A T2 HASTE anatomical image is provided for anatomical reference. Tables Captions Table 1. Parameters of the sequences used for T 1 and T 2 mapping. The same parameters were used in the phantom and in-vivo scans. Table 2. Cortex and medulla T 1 and T 2 mean and standard deviation in ms across the 10 healthy volunteers, as well as the coefficient of variation (CoV). . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted August 19, 2025. ; https://doi.org/10.1101/2025.08.17.25333832doi: medRxiv preprint