High Spatial Resolution ²³Na-MRI for Ischemic Brain Injury Detection

Abstract

High spatial resolution sodium (²³Na) imaging of brain lesions remains challenging due to the intrinsically low signal-to-noise ratio (SNR) of ²³Na-MRI compared with conventional proton (¹H) MRI. In this preclinical mouse study, we developed a high-resolution ²³Na-MRI platform based on a 14 T preclinical scanner using a dual-tuned head-implanted RF coil. This implanted coil configuration enabled longitudinal multimodal imaging in the same animal, including 1¹H-based T₂-weighted anatomical and diffusion-weighted imaging (DWI), and high-resolution ²³Na-MRI. To evaluate the capability of this approach for detecting acute ischemic injury, ischemic stroke was induced in mice using the middle cerebral artery occlusion (MCAO) model. Brain-wide ²³Na-MRI was acquired within the first 4 hours following MCAO at an in-plane resolution of 300 × 300 µm². Hyperintense ²³Na-MRI signals were observed in the ischemic brain regions, presenting a contrast-to-noise ratio (CNR) that exceeded that of T₂-weighted images. Infarct regions can be well identified by elevated ²³Na-MRI signal, which is spatially consistent with regions of reduced apparent diffusion coefficient (ADC) values, indicating water restriction in the ischemic core. This work established implanted dual-tuned ¹H/²³Na RF coils as a powerful approach for high-resolution ²³Na-MRI readout, facilitating the early detection of ischemic brain injury.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s12975-026-01503-z.

Introduction

Proton-based magnetic resonance imaging (¹H-MRI) has provided critical insights into the spatiotemporal evolution of ischemic brain injury, with diffusion- and relaxation-based MRI contrasts serving as sensitive indicators of cytotoxic edema and tissue damage [1–3]. These MRI contrast mechanisms, however, primarily reflect secondary changes in water mobility and tissue microstructure rather than the underlying disruption of cellular homeostasis that underlies irreversible injury. Loss of transmembrane sodium gradients is directly involved in cerebral ischemia. The resulting intracellular sodium ion accumulation reflects impaired Na⁺/K⁺-ATPase activity, membrane depolarization, and breakdown of ionic compartmentalization, processes that are tightly linked to cellular viability [4]. Sodium MRI (²³Na-MRI), therefore, provides a biophysically meaningful sodium-weighted image contrast by directly mapping tissue sodium distributions in vivo [5–11]. Despite this conceptual advantage, the utility of ²³Na-MRI in ischemic stroke has been constrained by insufficient spatial resolution and sensitivity.

The intrinsically low sensitivity of ²³Na-MRI, approximately 20,000–30,000 times lower than that of ¹H-MRI [12, 13], severely limits signal-to-noise ratio (SNR) and achievable spatial resolution [14, 15]. Consequently, prior studies in rodent stroke models, typically performed at spatial resolutions of ≥ 1 × 1 × 1 mm³ (Supplementary Table 1), have primarily reported global or regional sodium signal changes, with limited ability to resolve heterogeneity within infarct cores or penumbral regions [16–20]. Even at ultra-high magnetic field strengths, e.g., 21 Tesla (T) [21, 22], achievable voxel size remains on the order 1 mm³ range, substantially larger than those routinely obtained with diffusion- or T₂-weighted ¹H-MRI. This disparity in spatial resolution complicates direct biophysical comparisons between sodium accumulation and established proton-based markers of ischemic injury.

In this study, we address these limitations by combining ultra-high-field MRI at 14 T with a dual-tuned, head-implanted radiofrequency (RF) coil optimized for ¹H/²³Na imaging in a mouse model of middle cerebral artery occlusion (MCAO). This platform enables high-resolution ²³Na-MR at submillimeter spatial resolution, allowing spatially resolved comparisons of sodium accumulation, diffusion restriction, and T₂ changes within ischemic tissue. By improving both sensitivity and spatial fidelity, our approach establishes a framework for interrogating sodium-based biophysical mechanisms of ischemic injury at a spatial scale previously accessible only with proton-based MRI.

Methods and Data Analysis

Animals

C57BL/6 mice (Charles River Laboratories, MA) were housed (group 3–4/cage) under a 12-h light/dark cycle with food and water ad libitum. They were maintained under a 12-hour light/dark cycle with unrestricted access to food and water. All animal procedures were performed in compliance with protocols approved by the Animal Care and Use Committee (IACUC) at Massachusetts General Hospital (MGH), and the animals were cared in accordance with the National Research Council’s Guide for the Care and Use of Laboratory Animals.

Experimental Workflow

The RF coil was head-implanted one week before the MCAO stroke surgery induction. Following stroke induction, the mice were maintained on a warming pad for 15 min and then positioned within a 3D-printed cradle for MRI scanning.

RF Coil Implantation

The general procedure for RF coil implantation is described previously [23]. Briefly, mice were anesthetized with isoflurane (2.5% for induction and 1%–1.25% for maintenance vaporized in 70% N₂/30% O₂). The mouse head was fixed in a stereotaxic device, and the craniotomy incision (the size of the RF ring) was exposed to position the coil on the skull. The cyanoacrylate glue and dental cement (Stoelting Co., Wood Dale, IL) were applied to secure the coil and to firmly cover the underlying brain.

The Focal Cerebral Ischemia Stroke Model

C57BL/6 male mice were subjected to either permanent or 60-minute transient MCAO (pMCAO or tMCAO). Surgery was performed during the inactive phase/light cycle (7:00 to 19:00). General surgical preparation was performed as described previously [24, 25]. MCAO was performed under isoflurane anesthesia (2.5-3% induction, 1.5-2% maintenance in 70% N₂O/30% O₂), and body temperature was maintained at 37 ± 0.5 °C using a heating pad. A laser Doppler flowmetry (LDF) probe was attached to the right middle cerebral artery (MCA) territory (2 mm posterior, 2 mm lateral from the bregma) to monitor cerebral blood flow (CBF). The ventral midline neck was incised, and the right common carotid artery (CCA) bifurcation was exposed without damaging the nerves. After ligation of CCA, the external carotid artery (ECA) was cut in the middle of both ligations. Then, a nylon monofilament (602212PK10Re, Doccol Corporation, Sharon, MA, USA) was inserted into the ECA and advanced to the MCA origin through the internal carotid artery (ICA). In the tMCAO group, the CCA was opened after monitoring CBF persistence, and then the mouse was kept under anesthesia during occlusion periods. At 60 min after tMCAO, the filament was removed to restore CBF as reperfusion. For pMCAO, the CCA remained closed and the filament was not removed until sacrifice. Immediately after surgery, 10 ml/kg of saline was administered intraperitoneally (IP) to prevent dehydration. Mice were then transferred to a warm recovery chamber and scanned by MRI for 4 h after surgery. After scanning, mice were sacrificed to analyze infarct volume. In total, 7 mice were used, divided into 3 tMCAO and 4 pMCAO.

¹H-MRI and ²³Na-MRI Acquisiton

All images were acquired with a 14 T horizontal bore magnet (Magnex Sci, UK), interfaced to a Bruker AV-Neo console running ParaVision 360 V.3.3 (Bruker, BioSpin GmbH, Germany), and equipped with a 6 cm microimaging gradient set, capable of providing 1.2 T/m (Resonance Research, MA). A diameter of ~ 6 mm dual-tuned transceiver surface implantable coil (MRIBOT LLC, MA) was used to acquire MRI images. The dual-tuned surface coil was tuned and matched outside of the magnet bore. The ¹H-MRI scans were acquired to verify the correct animal positioning and shimming inside the magnet. The T₂-weighted (T₂W) rapid acquisition with relaxation enhancement (RARE) and diffusion-weighted imaging (DWI) scans were acquired first. Magnetic field B₀ homogeneity was optimized first by automatic shimming from ¹H-MRI acquisition and then used for ²³Na-MRI shimming. For the subsequent ²³Na-MRI scan, the optimized RF power for the ²³Na-MRI scans was measured using the AdjRefPowX report provided by the Bruker sequence (Supplementary Fig. 1).

The T₂W anatomical, DWI, and ²³Na-MRI scans were acquired with the same FOV (14.4 × 9.6 × 8 mm³, slice thickness = 0.5 mm) unless otherwise indicated. A modified ultrafast FLASH-based fMRI 2D line-scanning pulse sequence was applied for ²³Na-MRI acquisition [26]. The pulse sequence scripts and binary code implementations of the 2D line-scanning method are available at our shared source: https://tnnc.mgh.harvard.edu/method. The reconstructed images were analyzed without correction for the spatial sensitivity profile of the surface coil. The parameters for ²³Na-MRI are applied following: TE ~ = 1.5 ms, TR = 10 ms, matrix = 48 × 32 × 16, resolution = 300 × 300 × 500 µm³, unless otherwise indicated. For T₂W, a RARE sequence was implemented to acquire anatomical images with the same geometry. TE = 12 ms, TR = 1500 ms, RARE factor = 6, averages = 16, matrix = 96 × 64 × 16, resolution = 150 × 150 × 500 µm³. For DWI, TE = 14 ms, TR = 2500 ms, b values of 0, 50, 100, 150, 250, 400, and 600 s/mm² were applied and 96 × 64 × 16, resolution = 150 × 150 × 500 µm³. During the scanning, the mice were anesthetized with isoflurane 1%–1.25%. The body temperature was maintained at 37 °C by blowing warm air through the bore and recording with a rectal probe (Small Animal Monitoring Model 1030, SA Instruments, Inc., NY, USA).

Brain Coronal Section Staining for Ischemic Area

To evaluate the tissue outcome, mice brains were cut at coronal section and stained by 2,3,5-triphenyl-tetrazolium chloride (TTC) or Hematoxylin and Eosin (H&E).

TTC Staining

Mice were euthanized by cervical dislocation under deep anesthesia (5% isoflurane). Brains were carefully removed and sectioned into 1 mm-thick coronal slices and immersed in 2% of 2,3,5-triphenyl-tetrazolium chloride (TTC; Sigma-Aldrich, Inc., St. Louis, MO, USA) saline until intact tissue was sufficiently stained.

H&E Staining

Mice were deep anesthetized with 5% isoflurane and systemically perfused with saline for 2 min via the left ventricle. Brains were carefully removed, frozen in isopentane on dry ice, and stored at -80 degrees. Twenty micrometer-thick cryosections were collected every 1 cm on ionized glass slides and stained with hematoxylin and eosin (H&E; Poly Scientific R&D Corp., Bay Shore, NY, USA).

SNR and CNR Measurements

The SNR was computed by averaging the signal from the hyperintense ROI voxel delineating the center of the ischemic lesion regions and dividing by the standard deviation of the voxel from the background outside the brain region (Fig. 3A). The contralateral ROI voxels were positioned on the corresponding contralateral normal area with the same ROI size. The CNR was computed as the difference between the mean signal intensities of the hyperintense ²³Na area and the corresponding contralateral ROIs, divided by the standard deviation of the background region [27]. For SNR and CNR calculations, the RARE data were downsampled to the same resolution as the ²³Na-MRI map. The apparent diffusion coefficient (ADC) maps were generated from the nonlinear least-squares fit through all b values for each slice acquired by DWI. The spatial distribution profile was determined by the hyperintense ²³Na infarct area from the acquired ADC and ²³Na images. For the group analysis, a Wilcoxon test was performed. Data processing and analysis were performed using Analysis of Functional NeuroImages (AFNI) (NIH, Bethesda, USA) and MATLAB (MathWorks, Natick, USA). The error bars in all figures represent the mean ± standard error of the mean (SEM) unless otherwise indicated. The threshold for statistical significance was set at p < 0.05.

Fig. 3.

Quantitative comparison of ischemic lesion associated signal in ²³Na-MRI and T₂-weighted MRI. (A) Representative images acquired following MCAO demonstrated a hyperintense ²³Na signal within the lesioned brain areas compared with T₂-weighted (T₂W) rapid acquisition with relaxation enhancement (RARE) images. ROIs were defined within the lesioned area (red), the contralateral ROI (green), and the background outside the brain (light blue) for SNR and CNR calculation. Imaging parameters were as follows: T₂W RARE images, spatial resolution = 150 × 150 × 500 µm³, matrix size = 96 × 64 × 16, number of averages = 32, acquisition time = 16.01 min; ²³Na-MRI images, spatial resolution = 300 × 300 × 500 µm³, matrix size = 48 × 32 × 16, number of averages = 256, mean acquisition time = 93.18 min. Scale bars represent 1 mm. (B) The lesion SNR measured using ²³Na-MRI signals (26.34 ± 5.71) remained lower than that measured using T₂W RARE images (40.10 ± 3.91). No significant difference in SNR was observed between the infarct region and contralateral regions in the T₂W images, whereas a significant difference was detected in the ²³Na-MRI images (p = 0.03). The CNR between the infarct and contralateral regions measured by ²³Na-MRI images (15.80 ± 3.20) was significantly higher than that in the T₂W (4.47 ± 1.59, p = 0.04). Data are presented as the mean ± SEM (n = 5 mice). Statistical comparisons were performed using paired Wilcoxon signed-rank tests

Results

Head-Implanted RF Coils Enable High-Resolution ²³Na-MRI of the Mouse Brain

Building on our previously developed head-implanted RF coil technology [23, 28], we implanted a lightweight (< 2.5 g) dual-tuned RF coil directly onto the mouse skull (Fig. 1A). The coil was tuned to both ¹H and ²³Na resonance frequencies for 14 T preclinical MRI scanner (Fig. 1B), enabling switching between proton- and sodium-based acquisitions. This configuration allowed B₀ shimming and slice selection to be optimized using high-SNR ¹H-MRI and subsequently applied to ²³Na-MRI acquisitions, providing inherently co-registered proton and sodium images without geometric mismatch.

Fig. 1.

High-resolution multimodal ¹H-MRI and ²³Na-MRI of ischemic brain injury in a mouse model of middle cerebral artery occlusion (MCAO). (A) A dual-tuned RF coil was implanted onto the mouse skull to acquire ¹H and ²³Na-MRI signals. (B) The dual-tuned RF coil operated at respective ¹H and ²³Na resonance frequencies on a 14 T scanner, enabling sequential acquisition of both nuclei from the same animal. (C) Representative ¹H T₂-weighted (T₂W) anatomical images and high-resolution ²³Na-MRI from the same MCAO ischemic stroke mouse. Two ²³Na-MRI imaging protocols were used: 150 × 150 × 500 µm³ spatial resolution with an acquisition time of 115 min, 128 averages, and a matrix size of 80 × 64 × 24; and 300 × 300 × 300 µm³ isotropic resolution with an acquisition time of 128 min, 256 averages, and a matrix size of 48 × 32 × 40. The location and extent of ischemic infarction identified by MRI were validated using hematoxylin and eosin (H&E) staining

Using this platform, we successfully performed high-resolution ²³Na-MRI in mice subjected to MCAO, enabling visualization of ischemia-associated sodium accumulation. Representative images acquired at spatial resolutions of 150 × 150 × 500 μm³ (115 min acquisition) and 300 × 300 × 300 μm³ (128 min acquisition) revealed clear sodium hyperintensity in ischemic regions after stroke induction (Fig. 1C). These sodium-defined lesions spatially corresponded to infarct regions identified on T₂-weighted ¹H-MRI and were independently confirmed by hematoxylin and eosin (H&E) staining. Notably, the achieved sodium imaging resolution approached that of conventional proton-based anatomical imaging, enabling direct visual correspondence between modalities.

To evaluate the temporal evolution of ²³Na accumulation during the acute phase of ischemia, we performed serial ²³Na-MRI in a representative MCAO mouse at an in-plane resolution of 300 × 300 μm² (Fig. 2A). Dynamic imaging revealed a progressive increase in ²³Na signal intensity within the infarct region over time, with images acquired at ~18.7-minute intervals following stroke induction (Fig. 2B). Quantitative analysis demonstrated a ~ 112% increase in ²³Na signal within the infarct core over a 2.5-hour measurement period, while sodium signal in the contralateral hemisphere remained stable (Fig. 2C). These results demonstrate the feasibility of high-resolution ²³Na-MRI for mapping the spatial and temporal progression of brain lesions with submillimeter precision during the early stages of ischemic stroke.

Fig. 2.

Temporal dynamics of the ²³Na-MRI signal following transient MCAO. (A) Representative coronal T₂W and ²³Na-MRI images acquired following the transient MCAO stroke induction. The selected coronal slices correspond to the locations shown in the three-dimensional T₂W and ²³Na-MRI datasets. (B) Representative ²³Na-MRI slices at successive time points beginning approximately 3 h after transient MCAO surgery. (C) Time course of the ²³Na-MRI signal measured within the ischemic lesion and the corresponding contralateral regions of interest (ROIs) shown in (B). ROIs were delineated based on the hyperintense regions identified in the ²³Na-MRI slices. Image parameters were as follows: 300 × 300 µm² in-plane resolution, 500 μm slice thickness, 40 × 32 × 16 matrix size, 24 averages per repetition

Quantitative Characterization of ²³Na Signal Enhancement in Ischemic Lesions

To quantify lesion-associated sodium signal changes, ²³Na-MRI was performed in mice within the first four hours following MCAO induction. Sodium images were acquired using a multi-slice 2D FLASH-based sequence with an in-plane resolution of 300 × 300 µm² and a slice thickness of 500 μm. Robust sodium hyperintensity was consistently observed in ischemic regions on ²³Na-MRI across animals (Fig. 3A). For comparison, T₂-weighted ¹H images were acquired using a RARE sequence with matched slice geometry (150 × 150 µm² in-plane resolution, 500 μm slice thickness).

Although the absolute SNR of ²³Na-MRI in lesion regions (26.34 ± 5.71) was lower than for ¹H-MRI based RARE images (40.10 ± 3.91), the contrast-to-noise ratio (CNR) between infarct and contralateral regions was substantially higher for ²³Na-MRI (15.80 ± 3.20) compared with T₂-weighted imaging (4.47 ± 1.59; Fig. 3B). These findings indicate that, despite intrinsic lower SNR, ²³Na-MRI provides superior sensitivity for detecting ischemia lesions compared with conventional ¹H based T₂-weighted MRI.

Comparison of ²³Na-MRI with Diffusion-Based ADC Mapping

To further validate the lesion specificity of sodium-based contrast, ²³Na-MRI was compared with diffusion-derived apparent diffusion coefficient (ADC) maps in the same animals. Both modalities consistently identified ischemic regions across multiple slices, with lesion locations confirmed by TTC staining (Fig. 4A). In all animals, the ischemic hemisphere showed a pronounced increase in ²³Na signal intensity compared with the contralateral hemisphere. Spatial correspondence between elevated sodium signal and reduced ADC values was observed within infarct cores (Fig. 4B), indicating overlap between sodium accumulation and diffusion restriction. Despite this overlap, group analysis of line profiles revealed that ²³Na-MRI produced a more spatially confined lesion profile with much better lesion to background contrast than ADC maps (Fig. 4C). Additionally, three-dimensional volumetric renderings further illustrate the enhanced lesion delineation provided by ²³Na-MRI relative to diffusion imaging (Supplementary Fig. 2). Together, these results demonstrate that high-resolution ²³Na-MRI offers robust sensitivity, superior lesion contrast, and high spatial specificity for characterizing cortical and subcortical brain lesions following ischemic stroke.

Fig. 4.

Comparison of ADC mapping and ²³Na-MRI in ischemic brain injuries. (A) Representative subcortical brain lesions were well identified in the ²³Na-MRI image and were also detectable in the diffusion-based ADC maps, despite variations in T₂-weighted maps and TTC staining. Image parameters were as follows: ²³Na-MRI, spatial resolution = 300 × 300 × 500 µm³, matrix size = 48 × 32 × 16, number of averages = 256, acquisition time = 93.18 min; ADC mapping, spatial resolution = 150 × 150 × 500 µm³, matrix size = 96 × 64 × 16, acquisition time = 16.8 min. Scale bars indicating 1 mm are shown in the figures. (B) Representative signal profiles across the infarct regions are shown in the ADC map and ²³Na image, normalized to their respective peak values. (C) Spatial distribution of ²³Na-MRI and ADC values aligned according to the peak ²³Na-MRI signal within the infarct core regions (mean ± SEM, n = 4 mice)

Discussion

Using ultra–high‑field MRI in combination with a head‑implanted, dual‑tuned RF coil, we demonstrate submillimeter‑resolution ²³Na‑MRI of the mouse brain in vivo and apply this capability to characterize ischemic injury following MCAO. This technical advance enables robust visualization of spatially confined sodium accumulation within infarcted tissue, presenting more salient lesion contrast when compared with conventional proton‑based T₂‑weighted imaging or diffusion‑derived ADC maps. Together, these findings establish high‑resolution ²³Na‑MRI as a sensitive and biophysically meaningful complement to standard ¹H‑MRI for preclinical stroke research.

Overcoming the Intrinsic SNR Limitations of ²³Na‑MRI

The principal challenge of in vivo ²³Na‑MRI is its intrinsically low SNR, which arises from several fundamental factors, including the lower gyromagnetic ratio of ²³Na relative to ¹H, its substantially lower tissue concentration compared with water protons, and quadrupolar relaxation that produces rapid biexponential T₂/T₂* decay [5, 29–31]. Together, these constraints have historically necessitated large voxel sizes and long acquisition times, limited spatial resolution, and obscured fine‑scale heterogeneity within ischemic lesions. In the present study, we addressed these limitations by combining ultra‑high magnetic field strength (14 T) with a skull‑mounted implantable RF coil optimized for dual‑nucleus (¹H/²³Na) operation. The 14 T magnetic field provides a substantial SNR gain through the supra-linear dependence of SNR on field strength B₀, scaling approximately as B₀^1.65 for phased array coils and B₀^2.0 for small surface coils [32, 33]. In addition, the implantable coil configuration provides several key advantages. First, its close proximity to the brain reduced susceptibility-related field distortion near air–tissue interfaces, thereby improving B₀ homogeneity and shimming performance at high field [34]. Second, the short distance between the receive element and cortical tissue substantially enhances local SNR. Third, dual-tuned RF operation enables seamless switching between proton and sodium imaging with identical slice geometry, facilitating precise multimodal co-registration.

Collectively, these features enabled ²³Na‑MRI at in-plane resolutions as high as 300 μm, approaching the spatial scale of conventional ¹H imaging and representing an order‑of‑magnitude improvement over preclinical sodium imaging studies, including those performed at higher magnetic field strength [16–20]. Nevertheless, implantable surface coils inherently exhibit limited penetration depth and B₁ inhomogeneity, leading to signal attenuation in deeper brain structures. Despite these constraints, we demonstrate reliable detection of subcortical ischemic lesions, underscoring the feasibility of this approach for whole‑brain sodium mapping in mice. Future RF sensing developments, including figure-8 coil geometries [23, 35, 36], multi‑element cryogenic arrays [37– 39], and wireless implantable RF coils with built-in varactor-based preamplifier [40, 41], could further improve sensitivity, depth coverage, and quantitative accuracy.

High‑Resolution ²³Na‑MRI Provides Superior Lesion Contrast

A key observation of this work is that high‑resolution ²³Na‑MRI yields markedly higher lesion-based CNR than conventional T₂‑weighted ¹H‑MRI despite its lower absolute SNR. In human studies, the ²³Na signal elevation within ischemic lesions is reported to be ~ 50% relative to the contralateral tissue [8, 14, 42, 43], whereas preclinical models exhibit an increase approaching ~ 100% [44]. By improving SNR, we were able to accurately delineate the ischemic injury volume (Fig. 1 & Supplementary Fig. 2) and detect a time-dependent ~ 112% elevation of ²³Na concentration within the infarct core (Fig. 2C). Also, our dynamic measurements demonstrate a gradual, time‑dependent increase in ²³Na signal within the infarct core during the acute post‑ischemic period, highlighting the potential of ²³Na-MRI to track lesion evolution rather than simply providing a static assessment of tissue damage. This enhanced CNR of²³Na-MRI reflects the strong biophysical coupling between sodium accumulation and the loss of cellular ionic homeostasis that defines irreversible ischemic injury (Fig. 3). By improving spatial resolution and reducing partial‑volume effects, we were able to delineate infarct boundaries more precisely and resolve heterogeneous sodium elevations within both cortical and subcortical regions.

In high-resolution ²³Na-MRI, voxel size, CNR, and acquisition time are tightly coupled because of the intrinsically limited SNR of sodium imaging. Smaller voxels improve spatial resolution, reduce partial-volume effects, and allow more precise delineation of infarct boundaries and heterogeneous sodium accumulation. However, reducing voxel volume also lowers SNR and lesion CNR, which must be compensated by signal averaging, higher field strength, improved RF sensitivity, or longer scan time. This tradeoff is illustrated by our parallel dynamic ²³Na-MRI study, which was designed to detect millisecond-scale ²³Na signal changes during neuronal activation [56]. In that work, the temporal resolution was prioritized with a short TE/TR of 1/10 ms, and each dynamic trial was acquired within approximately 1–2 min. To preserve sufficient SNR under this temporal constraint, relatively large mouse-brain voxels of 600 × 600 × 800 µm³ were used, and extensive averaging was required; the SNR increased with averaging and approached approximately 45 after ~ 200 repetitions. This example highlights a general constraint in ²³Na-MRI that improvements in temporal or spatial resolution are balanced with SNR, CNR, and practical acquisition time. For this MCAO study, the acquisition was optimized for high-spatial-resolution mapping of sodium accumulation. As summarized in Supplementary Table 1, prior preclinical ²³Na-MRI studies of ischemic injury have used a range of field strengths, voxel sizes, scan times, and acquisition strategies, reflecting different compromises among resolution, sensitivity, and temporal efficiency. Within this context, the high magnetic field strength and head-implanted dual-tuned RF coil used here partially mitigate the SNR penalty of submillimeter imaging, enabling sufficient lesion CNR within a practical acquisition time. Thus, the selected parameters provide a favorable balance for detecting acute ischemic sodium accumulation and delineating lesion heterogeneity in the mouse brain.

Relationship Between Sodium Accumulation and Diffusion Restriction

Comparison with diffusion‑derived ADC maps revealed substantial spatial overlap between regions of elevated sodium signal and reduced ADC, consistent with prior observations that both contrasts reflect ischemia‑induced cellular dysfunction [45, 48]. Nevertheless, line‑profile and volumetric analyses showed that ²³Na‑MRI produced more spatially lineated lesion boundaries and higher lesion-to-background contrast than ADC maps. These differences likely reflect distinct underlying biophysical mechanisms. ADC reductions occur rapidly following ischemia onset due to cytotoxic edema and water redistribution [49, 50], whereas sodium accumulation develops progressively as a result of breakdown of membrane integrity, Na⁺/K⁺‑ATPase dysfunction, and loss of ionic gradients [44]. Therefore, these differing temporal dynamics suggest that combined ADC and ²³Na‑MRI measurements could provide complementary information regarding tissue states and injury progression. In principle, regions exhibiting diffusion restriction without substantial sodium elevation may correspond to metabolically compromised but potentially salvageable penumbral tissue, whereas regions with pronounced sodium accumulation likely represent irreversibly damaged infarct core (Fig. 4). Comparable to the conventional diffusion/perfusion weighted imaging (DWI–PWI) mismatch framework [51–54], a mismatch between ADC‑defined lesions and ²³Na‑defined sodium accumulation could therefore offer an alternative strategy for identifying penumbral tissue and estimating ischemic progression [14, 18, 42, 43].

Biological Interpretation and Limitations

Because ²³Na is a quadrupolar nucleus with spin 3/2, its NMR visibility in biological tissue is influenced by interactions with macromolecules, membranes, and restricted intracellular and extracellular microenvironments. These interactions can produce rapid biexponential transverse relaxation, such that sodium ions with very short relaxation components may be only partially detected at the echo times used in this study. Thus, the measured ²³Na-MRI signal should not be interpreted as a direct measure of total tissue ²³Na concentration, but rather as a relaxation-weighted signal from the NMR-visible sodium pool [31, 55]. Our prior in vivo ²³Na-MRI T₂ relaxation measurements demonstrated biexponential signal decay consistent with quadrupolar relaxation behavior reported in brain tissue [56, 57]. In this context, the value of the present work is that it enables high spatial resolution, relaxation-weighted ²³Na-MRI detection of sodium signal alterations during ischemic stroke progression in preclinical models.

Accordingly, the echo time (TE) of ~ 1.5 ms used in this study was selected based on our prior in vivo ²³Na relaxation characterization at 14 T and the known relaxation behavior of sodium in brain tissue [56]. We performed in vivo T₂ mapping of the mouse cortex and observed biexponential T₂ values (T_2Short = 1.8 ms and T_2Long = 11.7 ms), consistent with the quadrupolar relaxation properties of ²³Na. Based on these measurements, an ultrashort TE of 1.5 ms was chosen to preserve a substantial fraction of the rapidly decaying ²³Na signal while remaining compatible with the gradient performance, spatial resolution, and SNR requirements of the imaging protocol. These choices are also consistent with prior preclinical ²³Na-MRI studies, which have commonly used short echo times of approximately 0.3–3 ms to reduce signal loss from rapidly decaying sodium components, together with relatively short repetition times of approximately 8–30 ms to improve acquisition efficiency [57–61].

The increased ²³Na signal following MCAO most likely reflects disruption of ionic homeostasis caused by ischemic energy failure. ATP depletion impairs Na⁺/K⁺-ATPase activity, resulting in membrane depolarization and intracellular ²³Na accumulation [62]. However, conventional ²³Na-MRI does not directly distinguish intracellular from extracellular ²³Na. The measured signal may therefore reflect several concurrent processes, including intracellular ²³Na accumulation, changes in extracellular ²³Na concentration and volume fraction [42, 63, 64], cytotoxic and vasogenic edema, altered paravascular fluid transport [65], and regional perfusion heterogeneity [6, 66]. Because intracellular ²³Na accumulation is among the earliest consequences of ischemic energy failure [67, 68], regions exhibiting greater ²³Na signal elevation may represent more severe ionic and metabolic disruption and may be more likely to progress toward irreversible infarction. More modest elevations may reflect heterogeneous tissue states, potentially including metabolically compromised but salvageable tissue within the ischemic penumbra. This interpretation is supported by the close spatial correspondence observed in our study among increased ²³Na-MRI signal, reduced ADC, and histologically confirmed infarction. Nevertheless, because our measurements were not quantitatively calibrated or compartment-specific, the magnitude of the ²³Na-MRI signal alone cannot definitively determine tissue viability or distinguish infarct core from penumbra.

Taken together, these findings indicate that high-resolution ²³Na-MRI provides a noninvasive spatial marker of disrupted sodium homeostasis during ischemic stroke progression and offers biological information complementary to the water-mobility and tissue-microstructural changes detected by conventional ¹H-MRI. Separating intracellular and extracellular ²³Na contributions remains an important challenge in developing ²³Na-MRI as a quantitative biomarker of tissue viability. Advanced acquisition and MRI contrast mechanisms, including multiple‑quantum filtering [69, 70], and optimized radial or ultrashort‑echo trajectories [71–73], may offer potential avenues to increase sensitivity to intracellular sodium and facilitate separation of compartment-specific contributions. Integration of these techniques with the dual‑tuned RF platform described here could substantially enhance the specificity of sodium‑based imaging biomarkers.

Conclusion

In summary, this present study demonstrates that technical barriers that have long limited ²³Na‑MRI can be overcome in preclinical settings, enabling spatial resolutions approaching those of conventional proton MRI. By directly probing ionic dysregulation, high‑resolution sodium imaging provides information that is fundamentally distinct from hemodynamic, diffusion‑based, or relaxation‑based contrasts. In this study, T₂-weighted imaging is interpreted as reflecting edema and late-stage infarction, whereas ADC/DWI is used to identify acute to subacute ischemic infarction. In contrast, ²³Na-MRI provides complementary information that is not fully captured by conventional water-related structural and diffusion changes. When combined with diffusion and perfusion imaging, ²³Na-MRI has the potential to refine the assessment of the infarct core and penumbra, improve estimates of stroke onset and progression, and ultimately enhance the translational relevance of preclinical stroke models.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contributions

Yuanyuan Jiang, Data curation, Software, Formal analysis, Validation, Investigation, Methodology, Funding acquisition, Writing–original draft; Xiaoqing Alice Zhou, Data curation, Formal analysis, Validation, Investigation, Methodology; Takahiko Imai, Data curation, Formal analysis, Validation, Methodology, Writing–original draft; Lydia Hawley, Methodology; David Hike, Methodology; Sohail Mohammed, Methodology; Grace Yu, Methodology; Cenk Ayata, Resources, Data curation, Supervision, Funding acquisition; David Chung, Resources, Data curation, Supervision, Funding acquisition, Investigation, Visualization, Writing–review and editing; Xin Yu, Conceptualization, Resources, Data curation, Supervision, Funding acquisition, Validation, Investigation, Visualization, Project administration, Writing– review and editing.

Data Availability

All data is included in the Source data files, raw data that support the findings of this study are available from the corresponding authors upon request.

Declarations

Competing interests

X.Y. is the co-founder of MRIBOT LLC. The other authors declare no competing interests in relation to this work.