Imaging Markers Derived From MRI-Based Automated Kidney Segmentation

An analysis of data from the German National Cohort (NAKO Gesundheitsstudie)

Background: Population-wide research on potential new imaging biomarkers of the kidney depends on accurate automated segmentation of the kidney and its compartments (cortex, medulla, and sinus).

Methods: We developed a robust deep-learning framework for kidney (sub-)segmentation based on a hierarchical, three-dimensional convolutional neural network (CNN) that was optimized for multi-scale problems of combined localization and segmentation. We applied the CNN to abdominal magnetic resonance images from the population-based German National Cohort (NAKO) study.

Results: There was good to excellent agreement between the model predictions and manual segmentations. The median values for the body-surface normalized total kidney, cortex, medulla, and sinus volumes of 9934 persons were 158, 115, 43, and 24 mL/m². Distributions of these markers are provided both for the overall study population and for a subgroup of persons without kidney disease or any associated conditions. Multivariable adjusted regression analyses revealed that diabetes, male sex, and a higher estimated glomerular filtration rate (eGFR) are important predictors of higher total and cortical volumes. Each increase of eGFR by one unit (i.e., 1 mL/min per 1.73 m² body surface area) was associated with a 0.98 mL/m² increase in total kidney volume, and this association was significant. Volumes were lower in persons with eGFR-defined chronic kidney disease.

Conclusion: The extraction of image-based biomarkers through CNN-based renal sub-segmentation using data from a population-based study yields reliable results, forming a solid foundation for future investigations.

Chronic kidney disease (CKD) is a commonly occurring illness defined as the sustained presence of abnormalities of kidney function or structure (1, 2). CKD affects an estimated 12.7% of adults in Germany (3). Routine clinical examinations and population-based research focus on evaluating kidney function, generally by estimating the glomerular filtration rate (GFR) from the concentration of serum creatinine, cystatin C, or both (4, 5). However, this approach fails to take account of the fact that serum creatinine levels rise only after loss of approximately 50% of the kidneys’ filtration function (6), and information on kidney structure and morphology is ignored. The identification of additional renal biomarkers, such as imaging-derived structural markers that also capture properties other than filtration, is therefore highly desirable.

Imaging techniques with and without the administration of contrast medium have emerged as valuable tools to quantify structural parameters of various organs, including the total volume of the kidney (TKV) and corticomedullary differentiation (7, 8, 9, 10, 11, 12, 13, 14, 15). While contrast-enhanced methods are optimal for corticomedullary differentiation, they have typically not been used in population-based studies. The generation and characterization of imaging-based kidney markers has recently become possible with the advent of plain imaging in large population studies (16). This requires automated segmentation of the kidneys, which is feasible with both more traditional machine learning techniques (17, 18, 19, 20) and modern deep learning approaches (21, 22, 23, 24, 25, 26, 27, 28, 29). The potential clinical applications of automated magnetic resonance imaging (MRI)-based segmentation of kidney markers include enhanced detection and monitoring of CKD. The relationship between the volumes of automatically segmented kidney compartments and estimated GFR (eGFR) remains, however, largely unexplored. As a first step, this observational study utilized several thousand abdominal MR images from the German National Cohort (NAKO) and modern deep learning approaches to quantify the volumes of the kidneys and their subcompartments, to derive the distributions of volumetric parameters in this large population-based sample, and to identify their relevant biochemical and clinical correlates (30, 31).

Methods

Study sample

The NAKO is a prospective population-based study of 205 415 participants aged 19–74 years at the time of enrolment between 2014 and 2019 (31). All participants provided written informed consent, and the responsible institutional review boards approved the study (30). The collection of the participant characteristics used in this project at the baseline visit is described in eMethods 1. The eGFR (mL/min/1.73 m²) was calculated using the CKD-EPI equation for serum creatinine and cystatin C (5).

Imaging, kidney volume determination, quality control, and validation

A subset of 30 861 participants underwent baseline whole-body MRI at 3.0 Tesla (Skyra, Siemens Healthineers, Erlangen, Germany) at one of five dedicated MRI centers that used identical equipment and protocols (16). Thorough image quality assessment ensured that the acquired images were complete and of high quality (16). For this study, the water and fat images derived from the whole kidney coverage 3D T1 Dixon scan of the thoracoabdominal region from the first release of NAKO imaging data (N = 11 207; eFigure 1) were used.

Figure 1

Example of automated kidney segmentation into cortical, medullary, and sinus compartments

The segmentation of the kidney into four compartments (cortex, medulla, sinus, cysts)—by means of a hierarchical, multi-scale U-net—is detailed in eMethods 2. Two experienced radiologists established the ground truth for model training through meticulous manual segmentation. The calculated volumes were normalized to body surface area (BSA; mL/m²) (32). As done previously, TKV was defined as the sum of cortex and medulla volumes of both kidneys (12). After thorough quality control, data from 9955 persons was available for statistical analysis, including 4471 persons with complete information on all evaluated variables and 2945 persons without CKD or commonly associated conditions (eFigure 1).

Agreement between manual and 3D convolutional neural network (CNN)-based segmentations was based on randomly selected images of 20 participants not contained in the original training set. Correlation coefficients (Spearman, Pearson) were classified as good (≥ 0.6), very good (≥ 0.8), and excellent (≥ 0.95). The Dice similarity coefficients and Bland–Altman plots were also assessed (33, 34).

Statistical analysis

The methods for all statistical analyses are described are detail in eMethods 3. Subgroups were compared using a two-sided t-test at a statistical significance level of 0.05. Univariable and multivariable association analyses were performed using linear regression, considering different hierarchical models.

Results

Segmentation of kidneys

The kidneys of 9955 participants were successfully segmented into cortex, medulla, and sinus (Figure 1). The agreement of TKV, cortex, medulla, and sinus volumes between manual readings and the corresponding CNN predictions were good to excellent, both between the two radiologists and between radiologist(s) and predictions (Figure 2, eFigure 2). For example, the Spearman correlation coefficient for TKV ranged from 0.94 for the inter-radiologist comparison to 0.99 (radiologist 1 vs. CNN, radiologist average vs. CNN). Bland–Altman plots showed that there were no major systematic biases between manual readings and predictions (eFigure 3), although manual readings yielded on average higher TKV, cortex, and medulla volumes but lower sinus volumes than predictions.

Figure 2

Statistics for reader agreement of volumetric kidney markers

Figure 3

Distribution of total kidney volume (TKV) in persons with and without eGFR-based kidney disease

Distributions and correlations of kidney imaging markers

Half of the study participants were men, the mean age was 52 years (SD = 11), and the mean body mass index (BMI) was 26.6 kg/m² (SD = 4.5) (Table). The proportions of persons reporting physician-diagnosed diabetes and kidney disease in a standardized questionnaire were 5% and 2%, respectively. The mean eGFR was 99.3 mL/min/1.73 m² (SD = 14.7).

Table

Basic characteristics of the 9955 study participants with high-quality imaging data

The distributions of BSA-normalized TKV and of cortex, medulla, and sinus volumes are shown for the overall sample and separately for men and women in the Table and in eTable 1: while men had on average greater TKV, cortex, and sinus volumes than women, this was not observed for medulla (eFigure 4). The median proportion of TKV attributed to cortex was 72.4% (interquartile range Q1–Q3: 70.1–75.3%). The sex-specific differences were more pronounced when volumes were not normalized to BSA (eTable 2), suggesting that differences in height and weight account for much of the sex difference in TKV. Consistent with previous research (35), we observed greater average BSA-normalized volumes of the left than the right kidneys (eFigure 5).

Figure 4

Distribution of total kidney volume (TKV) in persons with and without self-reported kidney disease

Percentiles of the distributions of kidney compartment volumes from a very large, population-based sample without frequently occurring conditions that affect kidney function have not been published to date. We therefore calculated these values in a subset of 2945 persons without diabetes mellitus, hypertension, gout, and self-reported or eGFR-defined CKD (< 60 mL/min/1.73 m²). The ranges and percentiles of both non-normalized and BSA-normalized TKV, cortex, medulla, and sinus volumes in eTable 3 can serve as a basis for future epidemiological and clinical imaging studies. The distribution of TKV in persons with CKD, defined as eGFR < 60, was clearly shifted towards lower values compared with those with eGFR ≥ 60 mL/min/1.73 m² (Figure 3). This shift was much less pronounced for self-reported kidney disease (Figure 4).

The correlations and bivariate associations of volumetric markers with anthropometry, diseases, and eGFR are described in detail in the eResults, eTable 4, and eFigures 6–11. Average BSA-normalized TKV was 161.4 (SD = 23.2) mL/m² in those without and 131.8 (SD = 25.1) mL/m² in those with eGFR-defined kidney disease (p < 0.001). The corresponding values were 161.3 (SD = 23.3) and 155.2 (SD = 25.6) mL/m² in those without and with self-reported kidney disease (p = 0.04).

Multivariable-adjusted associations of kidney markers with anthropometry, diseases, and eGFR

We performed multivariable regressions of BSA-normalized volumetric markers adjusted for demographic, anthropometric, and disease-related variables. A full model containing the predictors study center, age, sex, height, BMI, eGFR, diabetes mellitus, gout, and hypertension explained 36% of the variance of TKV; the strongest individual contribution came from eGFR, which alone explained 25% of the TKV variance (eTable 5). With respect to the presence of diabetes mellitus, gout, and hypertension, the strongest relationships were observed for the association of diabetes mellitus with BSA-normalized TKV and cortex volumes (eFigure 12). For example, persons with diabetes mellitus had on average 10 mL/m² higher BSA-normalized TKV than those without diabetes mellitus (p < 0.001, eTable 5). This is consistent with glomerular hyperfiltration in persons with diabetes mellitus, as the glomeruli are predominantly contained in the cortex. In the fully adjusted model, each 10 mL/min/1.73 m² higher eGFR was associated with 9.8 mL/m² higher BSA-normalized TKV (p < 0.001).

Multivariable-adjusted associations of eGFR with kidney volumes, anthropometry, and diseases

Lastly, we evaluated whether volumetric markers could explain variability in eGFR, the most commonly used marker of kidney function. The full model, which contained BSA-corrected TKV as well as the predictors listed above, explained 55% of the variance of eGFR (eTable 6). In four fully adjusted models for the four kidney compartments, the volume of each compartment showed a significant positive association (p < 0.001) with eGFR, whereas age, height, BMI, and the presence of gout and hypertension exhibited negative associations with eGFR.

Discussion

Our study of MRI-based kidney markers in 11 207 persons from the general population yielded the following findings:

• Automated segmentation of the kidneys and their compartments from thousands of abdominal MR images derived from a population-based study is feasible and efficient.
• Distributions of kidney compartment volumes could be calculated for the whole group and for a subset of persons without CKD or commonly associated conditions.
• The anthropometric, biochemical, and clinical correlates of volumetric markers that were identified were plausible in comparison with existing data from the literature.

Imaging markers differed significantly between persons with and without eGFR-based CKD. The presented method creates a foundation for further studies of imaging-based kidney markers and their potential clinical uses, such as improved detection and monitoring of kidney disease and better tailoring of treatments.

The segmentations from this study are in line with the published literature. For example, a paper based on contrast-enhanced computed tomography (CT) images from a cohort of 1344 potential kidney donors reported average TKV values of 269 cm³ in women and 325 cm³ in men (17), compared with 272 and 335 cm³ (1 cm³ = 1 mL) in our sample. Segmentation of MR images from 1852 Framingham Heart Study participants revealed TKV of 278 and 365 cm³, respectively (12). The relative cortex volume of 73% in our study agrees well with figures from other studies based on contrast-enhanced CT images. For example, a value of 72% can be deduced from two such studies (17, 29), whereas histological measures in mice yielded 66% (36). Likewise, the Dice indices for TKV for agreement between the radiologists and the prediction agreement were comparable with those from a previous analysis that applied a CNN to MRI data from a large cohort (26) and with those from two earlier CT-based studies (17, 29). However, the Dice indices for the medulla were lower than in an earlier CT-based study (29).

Most associations of anthropometric and clinical variables with imaging markers were also consistent with previous findings. Despite good overall agreement there is, however, an interesting difference between our study and the CT-based study by Wang and colleagues with regard to the univariable relation of cortical and medullary volumes to age (17). Whereas we found a significant inverse correlation between medullary volume and age that disappeared after adjustment for eGFR, Wang et al. reported significant positive correlations between medullary volume and age in both univariable and multivariable models (17). Conversely, an association between cortex volume and age became apparent only upon adjustment for eGFR in our analysis, whereas Wang et al. found that a negative univariable association between cortical volume and age turned into a positive association after adjustment for eGFR (17). Our results therefore do not confirm the increase of medullary volume with age that was observed in earlier studies (37, 38).

Potential explanations for the observed differences in the relationship of age to cortical and medullary volumes include the fact that our study additionally segmented sinus as a compartment. Moreover, age- or eGFR-dependent structural and functional changes may potentially affect contrast-enhanced CT and native MRI differently. (39, 40). These differences highlight the fact that imaging-derived volumes may depend on the properties and parameters of the imaging modalities employed; kidney function can influence corticomedullary differentiation and hence segmentation. In the absence of a ground truth based on histology, it may therefore be more appropriate to speak of “imaging-derived cortex volume” and “imaging-derived medullary volume” in order to emphasize that they represent estimates of the actual morphological properties that also depend on the respective imaging modality and the underlying functional properties.

Despite the pronounced and significant differences in TKV between persons with and without self-reported or GFR-based CKD, not all persons with CKD had a low kidney volume. Conversely, not all persons with low kidney volumes reported CKD or had low eGFR. Such discrepancies may be attributable to low awareness of CKD, or the presence of kidney diseases may have a variable effect on kidney size and filtration. Future studies should therefore address whether imaging techniques or biochemical kidney markers are better suited for diagnosing CKD and predicting its progression.

The strengths of our study include its sample size, the highly standardized image acquisition, processing, and quality control, and the population-based sampling. However, our analysis excluded persons with large kidney cysts and therefore does not permit inferences about specific nephropathies such as polycystic kidney disease. Moreover, the segmentation was based exclusively on structural fat/water images. Numerous other functional and quantitative modalities for kidney imaging exist, and future evaluations across modalities may provide additional insights (14, 15).

Further limitations lie in the fact that fine details of the medulla, such as single pyramids, cannot be resolved. Although contrast-enhanced CT or MR imaging is superior for sub-segmentation of kidney compartments, the values obtained in our study without contrast medium enhancement are plausible and align very closely with the findings of previous studies that used contrast-enhanced CT (17, 29). We therefore assume that the automated approach to kidney segmentation applied herein will also prove useful for the future identification and study of more advanced imaging features.

Summary

The calculated distributions of total cortical, medullary, and sinus volumes may serve as a reference for future studies, including their relation to molecular markers and prospective clinical outcomes. Imaging markers of kidney structure showed strong associations with biochemical markers of kidney function and differed clearly between persons with and without CKD in the general population.

Further authors

Jan Lipovsek, Maximilian Russe, Harald Horbach, Christopher L. Schlett, Matthias Nauck, Henry Völzke, Thomas Kröncke, Stefanie Bette, Hans-Ulrich Kauczor, Thomas Keil, Tobias Pischon, Iris M. Heid, Annette Peters, Thoralf Niendorf, Wolfgang Lieb, Fabian Bamberg, Martin Büchert, Wilfried Reichardt

Affiliations of the further authors

Acknowledgments

This project was conducted with data from the German National Cohort (NAKO) (www.nako.de). The NAKO is funded by the Federal Ministry of Education and Research (BMBF) [project funding reference numbers:

01ER1301A/B/C, 01ER1511D and 01ER1801A/B/C/D], federal states of Germany and the Helmholtz Association, the participating universities and the institutes of the Leibniz Association.

We thank all participants who took part in the NAKO study and the staff ofthis research initiative.

Funding

The work of EK, JL, HH, FB, MRu and AK was funded by SPP 2177 (3598/6–1 and 6–2 to AK, KE 2513/1–1 and 1–2 to EK, RU 1900/2–2 to MRu, BA 4233/10–1 and 10–2 to FB) of the German Research Foundation (DFG). The work of AK and PS was further funded by DFG Project ID 431984000 SFB 1453. The work of IMH was funded by DFG Project ID 387509280, SFB 1350 and Project ID 509149993, TRR 374.

Data sharing

The NAKO study data are available to scientists with approved research proposals. More information about data requests can be found at https://transfer.nako.de/transfer/index

Conflict of interest statement

MN has received payment from the NAKO study centers at the Institute for Clinical Chemistry and Laboratory Medicine, University of Greifswald for performing the NAKO immediate analyses.

TP, AL, and WL are executive committee members of the NAKO study.

WR has received financial support from the DFG (RADIOMICS Program).

The remaining authors declare that no conflict of interest exists.

Manuscript eceived on 7 August 2023, revised version accepted on 19 February 2024.

Corresponding author
Prof. Dr. med. Anna Köttgen, MD MPH

Institut für Genetische Epidemiologie

Universitätsklinikum Freiburg

Hugstetter Str. 49

79106 Freiburg, Germany

Cite this as:
Kellner E, Sekula P, Lipovsek J, Russe M, Horbach H, Schlett CL, Nauck M, Völzke H, Kröncke T, Bette S, Kauczor HU, Keil T, Pischon T, Heid IM, Peters A, Niendorf T, Lieb W, Bamberg F, Büchert M, Reichardt W, Reisert M, Köttgen A: Imaging markers derived from MRI-based automated kidney segmentation—an analysis of data from the German National Cohort (NAKO Gesundheitsstudie).
Dtsch Arztebl Int 2024; 121: 284–90. DOI: 10.3238/arztebl.m2024.0040

References

1. Eckardt KU, Coresh J, Devuyst O, et al.: Evolving importance of kidney disease: from subspecialty to global health burden. Lancet 2013; 382: 158–69

2. Kidney Disease: Improving Global Outcomes (KDIGO) CKD Work Group: KDIGO 2012 clinical practice guideline for the evaluation and management of chronic kidney disease. Kidney Int Suppl 2013; 3: 1–150.

3. Girndt M, Trocchi P, Scheidt-Nave C, Markau S, Stang A: The prevalence of renal failure—results from the German Health Interview and Examination Survey for Adults, 2008–2011 (DEGS1). Dtsch Arztebl Int 2016; 113: 85–91

4. Levey AS, Stevens LA, Schmid CH, et al.: A new equation to estimate glomerular filtration rate. Ann Intern Med 2009; 150: 604–12

5. Inker LA, Schmid CH, Tighiouart H, et al.: Estimating glomerular filtration rate from serum creatinine and cystatin C. N Engl J Med 2012; 367: 20–9

6. Walker HK, Hall WD, Hurst JW, (eds.): Clinical Methods: The history, physical, and laboratory examinations [Internet]. 3 rd edition Boston: Butterworths 1990. www.ncbi.nlm.nih.gov/books/NBK201/ (last accessed on 9 December 2022).

7. Johnson S, Rishi R, Andone A, et al.: Determinants and functional significance of renal parenchymal volume in adults. Clin J Am Soc Nephrol 2011; 6: 70–6

8. Gupta S, Singh AH, Shabbir A, Hahn PF, Harris G, Sahani D: Assessing renal parenchymal volume on unenhanced CT as a marker for predicting renal function in patients with chronic kidney disease. Acad Radiol 2012; 19: 654–60

9. Zöllner FG, Svarstad E, Munthe-Kaas AZ, Schad LR, Lundervold A, Rørvik J: Assessment of kidney volumes from MRI: acquisition and segmentation techniques. AJR Am J Roentgenol 2012; 199: 1060–9

10. Halleck F, Diederichs G, Koehlitz T, et al.: Volume matters: CT-based renal cortex volume measurement in the evaluation of living kidney donors. Transpl Int 2013; 26: 1208–16

11. Gillis KA, McComb C, Patel RK, et al.: Non-contrast renal magnetic resonance imaging to assess perfusion and corticomedullary differentiation in health and chronic Kidney disease. Nephron 2016; 133: 183–92

12. Roseman DA, Hwang SJ, Oyama-Manabe N, et al: Clinical associations of total kidney volume: the framingham heart study. Nephrol Dial Transplant 2017; 32: 1344–50

13. Gardan E, Jacquemont L, Perret C, et al.: Renal cortical volume: high correlation with pre- and post-operative renal function in living kidney donors. Eur J Radiol 2018; 99: 118–23

14. Hockings P, Laustsen C, Joles JA, Mark PB, Sourbron S: Special issue: Magnetic resonance imaging biomarkers of renal disease. Magnetic resonance materials in physics, biology and medicine. 2020; 33. link.springer.com/journal/10334/volumes-and-issues/33–1 (last accessed on 2 April 2024)

15. Bane O, Seeliger E, Cox E, et al.: Renal MRI: from nephron to NMR signal. J Magn Reson Imaging 2023; 58: 1660–79

16. Bamberg F, Kauczor HU, Weckbach S, et al.: Whole-body MR imaging in the German National Cohort: rationale, design, and technical background. Radiology 2015; 277: 206–20

17. Wang X, Vrtiska TJ, Avula RT, et al.: Age, kidney function, and risk factors associate differently with cortical and medullary volumes of the kidney. Kidney Int 2014; 85: 677–85

18. Will S, Martirosian P, Würslin C, Schick F: Automated segmentation and volumetric analysis of renal cortex, medulla, and pelvis based on non-contrast-enhanced T1- and T2-weighted MR images. MAGMA 2014; 27: 445–54

19. Gloger O, Tönnies K, Mensel B, Völzke H: Fully automatized renal parenchyma volumetry using a support vector machine based recognition system for subject-specific probability map generation in native MR volume data. Phys Med Biol 2015; 60: 8675–93

20. Jin C, Shi F, Xiang D, et al.: 3D Fast automatic segmentation of kidney based on modified AAM and random forest. IEEE Trans Med Imaging 2016; 35: 1395–407

21. Sharma K, Rupprecht C, Caroli A, et al.: Automatic segmentation of kidneys using deep learning for total kidney volume quantification in autosomal dominant polycystic kidney disease. Sci Rep 2017; 7: 2049

22. Kaur R, Juneja M: A survey of kidney segmentation techniques in CT images. CMIR 2018; 14: 238–50

23. Torres HR, Queirós S, Morais P, Oliveira B, Fonseca JC, Vilaça JL: Kidney segmentation in ultrasound, magnetic resonance and computed tomography images: a systematic review. Comput Methods Programs Biomed 2018; 157: 49–67

24. Couteaux V, Si-Mohamed S, Renard-Penna R, et al.: Kidney cortex segmentation in 2D CT with U-Nets ensemble aggregation. Diagn Interv Imaging 2019; 100: 211–7

25. da Cruz LB, Araújo JDL, Ferreira JL, et al.: Kidney segmentation from computed tomography images using deep neural network. Comput Biol Med 2020; 123: 103906

26. Langner T, Östling A, Maldonis L, et al.: Kidney segmentation in neck-to-knee body MRI of 40,000 UK biobank participants. Sci Rep 2020; 10: 20963

27. Daniel AJ, Buchanan CE, Allcock T, et al.: Automated renal segmentation in healthy and chronic kidney disease subjects using a convolutional neural network. Magn Reson Med 2021; 86: 1125–36

28. Zollner FG, Kocinski M, Hansen L, et al.: Kidney segmentation in renal magnetic resonance imaging—current status and prospects. IEEE Access 2021; 9: 71577–605

29. Korfiatis P, Denic A, Edwards ME, et al.: Automated segmentation of kidney cortex and medulla in CT images: a multisite evaluation study. J Am Soc Nephrol 2022; 33: 420–30

30. German National Cohort (GNC) Consortium: The German National Cohort: aims, study design and organization. Eur J Epidemiol 2014; 29: 371–82

31. Peters A, German National Cohort (NAKO) Consortium, Peters A, et al.: Framework and baseline examination of the German National Cohort (NAKO). Eur J Epidemiol 2022; 37: 1107–24

32. Mosteller RD: Simplified calculation of body-surface area. N Engl J Med 1987; 317: 1098

33. Bland JM, Altman DG: Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 1: 307–10

34. Zou KH, Warfield SK, Bharatha A, et al.: Statistical validation of image segmentation quality based on a spatial overlap index. Acad Radiol 2004; 11: 178–89

35. Kalucki SA, Lardi C, Garessus J, et al.: Reference values and sex differences in absolute and relative kidney size. A Swiss autopsy study. BMC Nephrol 2020; 21: 289

36. Zangeneh MM, Zangeneh A, Tahvilian R, Moradi R, Tehrani PR: Preclinical evaluation of hematoprotective and nephroprotective activities of Bellis perennis L aqueous extract on CCl4-induced renal injury in mice. Comp Clin Pathol 2018; 27: 1557–66

37. O’Neill WC: Structure, not just function. Kidney Int 2014; 85: 503–5

38. Denic A, Glassock RJ, Rule AD: Structural and functional changes with the aging kidney. Adv Chronic Kidney Dis 2016; 23: 19–28

39. Shang JN, Ren K, Wu WS, et al.: [Investigation of renal corticomedullary differentiation with age-related change on non-contrast-enhanced MRI]. Zhonghua Yi Xue Za Zhi 2016; 96: 1505–9.

40. Otsuka T, Kaneko Y, Sato Y, et al.: Kidney morphological parameters measured using noncontrast-enhanced steady-state free precession MRI with spatially selective inversion recovery pulse correlate with eGFR in patients with advanced CKD. Clin Exp Nephrol 2018; 22: 45–54