Counting glomeruli in human kidney specimens using ex vivo MRI without contrast agents

Abstract

Purpose

The total number of glomeruli is an important indicator of kidney function and susceptibility to disease. Glomeruli, however, are difficult to image using MRI without contrast enhancement. For this reason, previous studies have used cationic ferritin as a contrast agent to detect glomeruli in ex vivo and in vivo perfused kidneys. This study aimed to develop a new technique using MR microimaging at ultrahigh field, without contrast agents, to visualize and quantify glomeruli in human nephrectomy samples.

Methods

MRI was performed at 16.4 T using three‐dimensional T₁/${\mathrm{T}}_2^{*}$‐weighted gradient‐echo imaging at 30 to 200 μm isotropic resolution on cortical samples with volumes ranging from 155 to 1384 mm³. A segmentation pipeline based on shape recognition and clustering was developed to count glomeruli on MR images at 30‐μm isotropic resolution. Glomerular number and density were estimated after MRI using unbiased stereology, and glomerular distribution was visualized using a subset of samples imaged using confocal microscopy.

Results

Glomeruli were identified in the MR images as hypointense spheres. High spatial concordance between glomeruli identified by MRI, histology, and confocal microscopy was observed. Statistical analyses showed that glomerular counts from stereology and MR images were similar, with Pearson correlation and intraclass correlation coefficients of 0.98.

Conclusion

High‐resolution microimaging at 16.4 T combined with automated segmentation allowed identification and quantitation of human glomeruli ex vivo without the need for contrast agents. The results showed strong agreement with the gold‐standard stereological method for glomerular counting.

1. INTRODUCTION

Glomerular number and size predict the trajectory of kidney function in life. ¹ , ² They provide insight on structural and functional changes in aging and disease, and into the extent of nephrogenesis during kidney development. ³ Reduced nephron number has been associated with an increased risk of hypertension, cardiovascular, and chronic kidney disease. ² , ⁴ , ⁵ In humans, glomerular volume ranges from approximately 0.9 to 11 × 10⁶ μm³ (radius of 60–140 μm), ² with each adult kidney containing an average of 1 million nephrons. However, stereological studies identified a more than 12‐fold range in nephron number (from 210 000 to 2 400 000 per kidney) in “normal” individuals from different ethnic groups. ⁴ , ⁵ This high variability in glomerular morphology and number underscores the methodological importance of glomerular quantitation for research. The “gold standard” method for accurate and precise unbiased estimation of the total number of glomeruli in a kidney is a design‐based stereological method known as the physical disector/fractionator combination. ³ However, this method is not widely used because it requires access to a laboratory equipped for glycolmethacrylate processing and sectioning, and specialized equipment for simultaneous viewing of section pairs. It is labor‐intensive and time consuming; aside from the several days of sample preparation time, counting glomeruli using disectors in a single human kidney takes approximately 10–12 h. Recently, a range of techniques have been developed to clear tissue samples and stain components such as glomeruli for visualization and enumeration. ⁶ However, tissue preparation and immuno‐labeling takes days to weeks to complete; labeling is highly variable; and specialized microscopes are required to image samples thicker than a few hundred microns.

In this study, we aimed to develop an MRI‐based approach for ex vivo glomerular counting without contrast agents, using simpler handling and automated approaches, as an alternative to conventional histological techniques. It is expected that this new technique can be used to analyze nephrectomy/biopsy samples in a wide range of research and clinical applications. Glomerular counting using biopsy samples has been used to study a range of kidney diseases, such as glomerulosclerosis and fibrosis, and changes in glomerular number have been shown to be prognostic indicators of chronicity. ⁷ , ⁸ , ⁹ Furthermore, low nephron number for age, estimated from the examination of biopsy samples, predicts the risk of chronic kidney disease after living kidney donation. ¹⁰

Previous MRI studies using cationic ferritin to image glomeruli in human kidney (ex vivo) and animal models (ex vivo and in vivo) have been performed at 7 T. ¹¹ , ¹² , ¹³ , ¹⁴ In this study, we hypothesized that MR microimaging at ultrahigh field (16.4 T), without contrast agents, can be used to visualize glomeruli in human nephrectomy samples. The more than 2‐fold increase in the magnetic field from 7 T to 16.4 T is expected to significantly improve the signal‐to‐noise ratio, ¹⁵ which can be expended to achieve higher image spatial resolution. ¹⁶ Human nephrectomy samples were formalin‐fixed and imaged without contrast agents using multiresolution MRI at 16.4 T. Samples were then processed for glomerular quantitation and distribution using stereology and confocal imaging. Our results showed that glomeruli can be accurately segmented from MR images, yielding similar results to those obtained with stereology and confocal imaging.

2. METHODS

2.1. Tissue samples

Human kidney tissues were obtained with informed patient consent from the macroscopically/microscopically healthy portion of tumor nephrectomies, following approval by the Royal Brisbane and Women's Hospital Human Research Ethics Committee (2002/011). In this cross‐sectional study, the eligibility criteria were adults (18 years or older) able to give informed consent undergoing nephrectomy at the Royal Brisbane and Women's Hospital. Participants follow up was not recorded for this study. Tissue specimens were classified as “control/healthy” based on histopathological assessment of the nontumor kidney parenchyma by a kidney histopathologist blinded to the experimental procedure to avoid bias. Glomerular quantitation was performed using MRI and stereology on six samples from 4 participants (Table S1).

2.2. MR microimaging at 16.4 T

Tissue samples were obtained from whole‐kidney post‐nephrectomy without flushing the blood, fixed in formalin for 24 h and washed with saline for 3 days at 4°C before scanning. Samples were scanned at room temperature using a Bruker Biospec 16.4T MRI at the Center for Advanced Imaging, the University of Queensland. This scanner is equipped with a Micro2.5 gradient (1.5 T/m at 60 A). Microimaging coils appropriate to the size of the samples were used, these included a 15 × 30 mm surface coil and a 15‐mm and a 10‐mm volume coil. Samples were either placed in saline or wrapped with paraffin film to minimize air artifacts. MRI was performed using a three‐dimensional (3D) T₁/${\mathrm{T}}_2^{*}$‐weighted gradient‐echo sequence with repetition time of 150 ms, echo time of 6.2 ms, and flip angle of 60°. T₁/${\mathrm{T}}_2^{*}$ weighting describes a combination of predominant T₁ weighting and noticeable ${\mathrm{T}}_2^{*}$ weighting using these sequence parameters at 16.4 T.

Typically, the field of view was about 2.4 × 1.2 × 1.2 cm with the matrix sizes set to produce images at 30, 60, 100, and 200 μm 3D isotropic resolutions. The acquisition times were approximately 27 h (number of excitations [NEX] = 4), 3 h 20 mins (NEX = 2), 1 h (NEX = 2), and 9 mins (NEX = 1) for each of the image resolutions, respectively. A sine windowing function was applied to the k‐space data before the Fourier transform to reduce the noise in the 30‐μm images.

2.3. Glomerular counting using MRI

The MRI data were masked to remove the background and measure the sample volumes. In addition to reporting absolute glomerular counts, to account for the variability in sample volumes, results are also presented for glomerular density estimated by dividing the sample glomerular count by the corresponding sample volume.

Glomeruli appeared as small, hypointense structures (Figure 1A), commonly denoted as “blobs” in computer vision terminology. We adapted a blob detection method by Zhang et al., originally developed for contrast‐enhanced MRI and based on Laplacian of Gaussian and Hessian analysis, ¹⁷ to identify glomeruli in our contrast agent–free MR images.

FIGURE 1.

MRI glomerular segmentation. (A) Original MR image at 30‐μm isotropic resolution. (B) Laplacian of Gaussian kernel filtered image. Local maxima are shown in bright yellow, with surrounding tissue and background shown in blue. (C) Presegmentation using Hessian analysis. (D) Blob candidates (in yellow) with plausible size. (E) Final segmentation of glomeruli using Variational Bayesian Gaussian Mixture Model clustering. Glomerular structures are shown in green, and other non‐glomerular structures in red. Bars = 1 mm.

In this three‐step algorithm, blob candidates were first highlighted by convolving the MR image with a Laplacian of Gaussian (LoG) ¹⁸ kernel (Figure 1B). As the LoG operation is edge‐sensitive, an edge filter was applied to remove false detections along tissue borders. Second, blobs in the LoG‐enhanced image were delineated from surrounding tissue based on local convexity to filter out non‐glomerular structures (e.g., blood vessels; Figure 1C). Regional features were then extracted to characterize each blob candidate, and nonplausible candidates were filtered by size, using known glomerular size limits ² (Figure 1D). Finally, through exhaustive testing, glomerular features consisting of volume (V), second‐order structureness (S), and geometric ratio (R _B ) ¹⁹ were selected for postpruning using a Variational Bayesian Gaussian Mixture Model ²⁰ (Figure 1E). Detailed steps of the algorithm are provided in Supporting Information Method S1.

2.4. Validation of glomerular structures using histology

To compare glomerular visualization using MRI with a glomerular‐specific antibody staining, one sample was embedded in paraffin and sectioned using a microtome at 4 μm. Tissue sections were stained with an antibody for human podocalyxin, a sialoglycoprotein marker of glomerular podocytes (MAB1658l; R&D Systems), and horseradish peroxidase (Biocare Medical), counterstained with hematoxylin and mounted using DPX.

2.5. Glomerular counting using stereology

Glomerular numbers were independently counted on six samples using the physical disector/fractionator combination, a design‐based (unbiased) stereological approach. ²¹ , ²² Glomerular counts using MRI and stereology were performed independently in two separate laboratories, and the results from each laboratory were not revealed until all counting was completed. To enable comparison with MRI, histology sections were scanned using an Olympus Upright FL Microscope BX63 brightfield microscope at 4× magnification and constructed into a stack using Fiji TrakEM2 software. ²³ Further details on glomerular counting using stereology can be found in Supporting Information Method S2.

2.6. Glomerular distribution using confocal microscopy

Following MRI, samples B (20 × 8 × 4 mm), E (19 × 10 × 9 mm), and F (20 × 12 × 9 mm) were washed in phosphate buffer saline for 2 days. Glomerular structures were detected by confocal microscopy in cleared wholemount tissues as per Combes et al. with extended incubation times. ²⁴ Details on antibody staining can be found in Supporting Information Method S3.

Samples were imaged in a tiled 3D volume at 10× magnification on a 3i Marianas spinning disk confocal microscope. A total of 647 (NPHS1) and 488 (autofluorescence) channels were acquired for optical sections up to 2000 μm in depth, which is the working distance limit of the 10× objective. The NPHS1 antibody stained tissue up to 100 μm from the tissue border, whereas autofluorescence resolved glomerular structures for the first 2 mm of the confocal imaging volume, with the quality decreasing with the increased imaging depth. Glomeruli were segmented from the autofluorescence channel, supported by NPHS1 staining where available. Glomerular segmentation in the confocal images was performed using manual segmentation in the Imaris software package (Oxford Instruments) using the manual “Surfaces” function. Following confocal imaging, the whole samples were then processed and embedded in paraffin for sectioning and glomerular counting using stereology.

2.7. Statistical analysis

Statistical analyses were performed using GraphPad Prism v10.2.0, IBM SPSS v29.0.2.0, and G*Power v3.1.9.6 for MacOS. Normality of data was tested and confirmed using the Shapiro–Wilk tests. Pearson correlation, linear regression, Bland–Altman plots, and intraclass correlation coefficient were used to compare total glomerular counts between measurements obtained from 30‐μm MRI resolution and stereology. Measurement values are shown as mean ± standard deviation. The standard errors for estimates of the slope and Y‐intercept in linear regression were calculated. Further details on the statistical analysis can be found in Supporting Information Method S4.

3. RESULTS

3.1. Visualization and segmentation of glomeruli ex vivo using 16.4T MRI

In samples from healthy kidney tissue, glomeruli were recognized as hypointense elliptical structures in T₁/${\mathrm{T}}_2^{*}$‐weighted gradient‐echo images (Figure 2) and were distinguished from blood vessels by their shapes. Blood vessels were also hypointense but were elongated rather than spherical (Figure S1). Glomeruli were clearly visible at 30‐μm resolution (Figure 2A); they were less clear at 60‐μm and 100‐μm resolutions (Figure 2B,C, respectively) and could not be distinguished at 200‐μm resolution (Figure 2D). From this observation, images at 30‐μm resolution were selected for further analysis.

FIGURE 2.

The 16.4T human kidney ex vivo imaging. MR images were acquired at 30 μm (A), 60 μm (B), 100 μm (C), and 200 μm (D) image resolution. Yellow and red arrows in (A) and (B) indicate glomerular and blood vessel structures, respectively. Bars = 1 mm.

To assist comparison, cross sections of the MRI volume were coregistered with images of histological sections using sample edges and blood vessels as landmarks. Glomeruli identified on MRI had high spatial correlation with those identified using podocalyxin (Figure 3A,B) and peanut agglutinin (PNA) staining (Figure 3C,D). Small differences in the appearance of glomeruli were attributed to tissue shrinkage during paraffin embedding and differences in slice thickness and cutting angle used in the MRI and paraffin sections.

FIGURE 3.

Representative images of structural MRI at 16.4 T and coregistered with histological sections. (A,C) Formalin‐fixed human kidney sample imaged with MRI at 30‐μm isotropic resolution. Glomeruli are seen as hypointense globular structures in a three‐dimensional T₁/${\mathrm{T}}_2^{*}$‐weighted gradient‐echo MRI scan. (B) Podocalyxin immunostaining of sample in image (A). (D) PNA staining of sample in image (C). Glomeruli can be seen in the histology as brown globular structures. Numbers refer to corresponding glomeruli in paired A/B and C/D images. Scale bar = 1 mm. PNA. peanut agglutinin.

We next performed comparison between glomeruli segmented automatically on MR images with those identified in PNA‐stained histological sections. The location and distribution of segmented glomeruli in the MRI accurately reflected the distribution of glomeruli in matched PNA‐stained histological sections (Figure 4).

FIGURE 4.

Comparison between histology, MRI, and three‐dimensional glomerular counting. (A) Paraffin section with PNA staining. (B) The 30‐μm resolution T₁/${\mathrm{T}}_2^{*}$‐weighted MRI with approximately matching slice position with (A). (C,D) The 3.75× magnification of framed areas in (A) and (B), respectively. In (C), the glomeruli are highlighted in yellow. In (D), the segmented glomerular structures are shown in green. Scale bars = 1 mm. PNA. peanut agglutinin.

3.2. Comparison of glomerular distribution in MRI and confocal imaging

To compare the MRI data with another volumetric imaging method, three samples were cleared and imaged using confocal microscopy. Glomeruli were identified by their morphology in the 488/autofluorescence channel, as confirmed by colocalization with NPHS1, and manually segmented. Comparisons of segmented glomeruli from matched confocal and MRI volumes revealed a similar distribution throughout the tissue (Figure 5 and Figure S2).

FIGURE 5.

Glomerular quantification using tissue clearing and confocal imaging. (A) Three‐dimensional maximum intensity projection for a human kidney slice (Sample B) showing glomerular (podocyte) marker NPHS1 in magenta, autofluorescence in white, and fill from segmented glomerular objects in cyan. (B) Region from a single optical section showing the same markers as in (A). Boxed region enlarged with glomeruli outlined (dashed white lines) to illustrate three segmented glomeruli. (C) Three‐dimensional MRI reconstruction of glomeruli (cyan) of the same sample.

Estimates of glomerular number obtained with MRI and confocal imaging were not compared, because only part of the tissue volume could be visualized with confocal imaging. This was due to technical limitations related to the single illumination source and working distance of the detection objectives of the confocal microscope, which led to a loss of image quality and definition of structures with increasing depth in the cleared tissue sample. The optical clearing process also changed the sample size and tissue shape, impeding accurate estimation of the corresponding partial MRI volumes to use for quantitative analysis. With these limitations, we turned to stereology to compare glomerular number throughout the entire tissue imaged by MRI.

3.3. Comparison of glomerular counting by MRI and stereology

The glomerular total count and density estimated by MRI and stereology in six nephrectomy samples are given in Table S2. The mean estimate of total glomerular count (± standard deviation) for the six samples was 4213 ± 3612 using stereology compared with 3984 ± 2922 for MRI. Mean glomerular counts were not significantly different between stereology and MRI (paired t‐tests, p = 0.50). The large standard deviations reflect the wide range in sample volumes, from 155 to 1384 mm³ (mean: 718 ± 533 mm³). The range of cortical volumes was 133 to 1144 mm³ (mean: 589 ± 424 mm³), with the cortex comprising approximately 84 ± 3% of the sample volumes.

Glomerular density estimates, which account for differences in sample volume, were more consistent. Glomerular density estimates based on stereology and on MRI were respectively 5.6 ± 0.9 and 5.5 ± 0.5 glomeruli per mm³. This difference was not statistically significant (paired t‐test, p = 0.89; Figure S3).

Results of statistical analyses of correlation and agreement between MRI‐based estimates and stereology are provided in Table S3, and the linear regression and Bland–Altman plots are presented in Figure S4. Glomerular counts obtained using MRI produced very similar results to stereology, with a Pearson correlation r of 0.98 (p = 0.0008), a mean Bland–Altman ratio of 1.0 ± 0.2 and bias of −299 ± 988, and an average measures intraclass correlation coefficient of 0.98 (0.87–1.00, p < 0.001). For glomerular density, the bias of the two methods' Bland–Altman ratio was −0.1 ± 1.2.

4. DISCUSSION

Current MRI‐based methods for estimating nephron number rely on contrast agents, which pose challenges for clinical translation due to potential toxicity. ²⁵ , ²⁶ On the other hand, morphometric and stereological methods are either biased or time‐consuming. We have developed a microimaging protocol at 16.4 T to visualize glomeruli ex vivo without contrast agents, alongside an automated segmentation method for quantifying glomeruli from high‐resolution (30 μm) MR images. Our MRI‐based estimates of glomerular number and density showed excellent agreement with the gold‐standard stereological method. Notably, estimates of glomerular density can be combined with cortical volume measurements from clinical imaging to derive total glomerular number per kidney. ²⁷

Determination of nephron number in whole kidneys presents significant challenges. Current MRI‐based techniques for quantitation of glomerular number require perfusion with the contrast agent cationic ferritin, which binds to the glomerular basement membrane, into the intact kidney (e.g., through tail vein perfusion in rodents or catheterization of the renal artery in fresh human kidney ex vivo). ¹⁴ , ²⁸ , ²⁹ , ³⁰ However, in many forms of glomerular disease, the cationic ferritin appeared to leak through the glomerular filtration barrier and diffusely accumulate in tubules, affecting the accuracy of nephron detection. ¹² Problems with the cationic ferritin method are likely related to susceptibility and the acquisition protocol, which has to date been mostly with ${\mathrm{T}}_2^{*}$‐weighted MRI, but the technique should improve as acquisition and image processing are further developed. ¹⁴ , ³¹

By directly imaging glomeruli using MRI, the methods described in the present study alleviate these drawbacks and provide a simple, accurate, and automated method of nephron quantitation. Our technique does not require perfusion of contrast agents to an entire kidney and therefore can be used to study a broader range of clinical samples. Tissue samples only need to be fixed in 10% formalin overnight and can be refrigerated in saline until they are ready for MRI. Residual blood in the fixed tissue may contribute to the hypointense signal of the glomeruli on T₁/${\mathrm{T}}_2^{*}$‐weighted images at the ultrahigh field, as imaging using perfusion flushed porcine kidney samples did not produce visible glomeruli. The ease of sample preparation and storage is likely to increase glomerular quantification by enabling clinical teams to ship fixed human kidney samples to teams with expertise in MRI. The samples also remain intact post MRI, enabling their use in a variety of downstream analyses.

Reported image resolution and acquisition times of glomerular imaging with MRI using cationic ferritin contrast agents have been variable. For example, high‐resolution 17‐μm isotropic imaging of ex vivo human kidney at 7 T took about 11 h. ³² For previous studies in rodent kidneys at 9.4 T, varying scan times have been reported, from 30 min to 10 h, depending on method and resolution. ³³ , ³⁴ , ³⁵ , ³⁶ In comparison, our method eliminates the need for contrast agents, instead using ultrahigh‐field MRI at 16.4 T to achieve high‐resolution imaging at 30‐μm isotropic. The extended acquisition time (27h) is necessary to achieve adequate tissue contrast in the absence of contrast agent. When samples were incubated in a paramagnetic contrast agent solution before MRI (e.g., using 0.2% vol/vol Magnevist; Bayer), the MRI scan time was shortened by 75%, while maintaining image quality (see Figure S5).

The process of exporting the MRI data, image preprocessing and enumeration of glomeruli can be automated and typically took less than 1 h to complete, making this MRI‐based method much less demanding with respect to the manual handling time compared with histology and stereological analyses. The MRI method has greater flexibility for imaging thicker samples than optical imaging methods, but the latter method can report the distribution of additional kidney biomarkers using antibody staining. Microscopes optimized for cleared tissue imaging can accommodate larger samples but are not commonly available. We expect that techniques for glomerular counting will continue to be developed to overcome their particular limitations and that all the tools are likely to be useful under specific circumstances, constituting a set of complementary quantitative techniques.

In conclusion, we have developed a method to visualize and count glomeruli accurately and rapidly without the need for contrast agents using ultrahigh‐field MRI. Our technique employed simple handling and preparation, ensuring that samples remained intact and suitable for subsequent clinical analyses following imaging. It enables the determination of glomerular density from biopsies, thereby contributing to the assessment of nephron number and kidney function in settings such as incipient chronic kidney disease ²⁷ and the pre‐transplantation assessment of donor kidney quality. ³⁷ A single‐needle biopsy is insufficient for accurate estimation of total nephron number, due to spatial variation in glomerular density. ³¹ Our method could help to address this issue because of its potential to be scaled up to enable simultaneous quantitation of multiple core biopsies in a time and labor‐efficient manner. Future development of this MRI method will focus on characterization of the methodology in samples with pathology and implementation of deep neural networks to improve the accuracy of glomerular segmentation.

FUNDING INFORMATION

This work was funded by the Australian Research Council Discovery Project DP220102042.

CONFLICT OF INTEREST STATEMENT

Nothing to report.

Supporting information

Data S1. Supporting Information.

Kurniawan ND, Amar AJ, Cullen‐McEwen LA, et al. Counting glomeruli in human kidney specimens using ex vivo MRI without contrast agents. Magn Reson Med. 2025;94(6):2519‐2528. doi: 10.1002/mrm.30642

DATA AVAILABILITY STATEMENT

Data and code have been made freely available (open access). The segmentation algorithm in Python is available at https://github.com/1996Aurel/HLoG‐VBGMM.git, and MRI data are available at https://doi.org/10.48610/7da7765.