Clinical utility of BOLD-MRI in accurate diagnosis and prognostic evaluation of diabetic nephropathy: a prospective renal biopsy-based cohort study - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Insights Imaging . 2026 Apr 20;17:106. doi: 10.1186/s13244-026-02274-9 Search in PMC Search in PubMed View in NLM Catalog Add to search Clinical utility of BOLD-MRI in accurate diagnosis and prognostic evaluation of diabetic nephropathy: a prospective renal biopsy-based cohort study Qian Wang Qian Wang 1 Department of Nephrology, First Medical Center of Chinese PLA General Hospital, Nephrology Institute of the Chinese People’s Liberation Army, State Key Laboratory of Kidney Diseases, National Clinical Research Center for Kidney Diseases, Beijing Key Laboratory of Kidney Disease Research, Beijing, China 2 Medical School of Chinese PLA, Beijing, China Find articles by Qian Wang 1, 2, # , Shaopeng Zhou Shaopeng Zhou 2 Medical School of Chinese PLA, Beijing, China 3 Department of Radiology, First Medical Center of Chinese PLA General Hospital, Beijing, China Find articles by Shaopeng Zhou 2, 3, # , Yue Niu Yue Niu 1 Department of Nephrology, First Medical Center of Chinese PLA General Hospital, Nephrology Institute of the Chinese People’s Liberation Army, State Key Laboratory of Kidney Diseases, National Clinical Research Center for Kidney Diseases, Beijing Key Laboratory of Kidney Disease Research, Beijing, China Find articles by Yue Niu 1, # , Chaobo Li Chaobo Li 2 Medical School of Chinese PLA, Beijing, China 3 Department of Radiology, First Medical Center of Chinese PLA General Hospital, Beijing, China Find articles by Chaobo Li 2, 3 , Qiang Lyu Qiang Lyu 1 Department of Nephrology, First Medical Center of Chinese PLA General Hospital, Nephrology Institute of the Chinese People’s Liberation Army, State Key Laboratory of Kidney Diseases, National Clinical Research Center for Kidney Diseases, Beijing Key Laboratory of Kidney Disease Research, Beijing, China Find articles by Qiang Lyu 1 , Pu Chen Pu Chen 1 Department of Nephrology, First Medical Center of Chinese PLA General Hospital, Nephrology Institute of the Chinese People’s Liberation Army, State Key Laboratory of Kidney Diseases, National Clinical Research Center for Kidney Diseases, Beijing Key Laboratory of Kidney Disease Research, Beijing, China Find articles by Pu Chen 1 , Xiaojing Zhang Xiaojing Zhang 3 Department of Radiology, First Medical Center of Chinese PLA General Hospital, Beijing, China Find articles by Xiaojing Zhang 3 , Lizhi Xie Lizhi Xie 4 GE Healthcare, MR Research China, Beijing, China Find articles by Lizhi Xie 4 , Wanjun Shen Wanjun Shen 1 Department of Nephrology, First Medical Center of Chinese PLA General Hospital, Nephrology Institute of the Chinese People’s Liberation Army, State Key Laboratory of Kidney Diseases, National Clinical Research Center for Kidney Diseases, Beijing Key Laboratory of Kidney Disease Research, Beijing, China Find articles by Wanjun Shen 1 , Yong Wang Yong Wang 1 Department of Nephrology, First Medical Center of Chinese PLA General Hospital, Nephrology Institute of the Chinese People’s Liberation Army, State Key Laboratory of Kidney Diseases, National Clinical Research Center for Kidney Diseases, Beijing Key Laboratory of Kidney Disease Research, Beijing, China Find articles by Yong Wang 1 , Xueying Cao Xueying Cao 1 Department of Nephrology, First Medical Center of Chinese PLA General Hospital, Nephrology Institute of the Chinese People’s Liberation Army, State Key Laboratory of Kidney Diseases, National Clinical Research Center for Kidney Diseases, Beijing Key Laboratory of Kidney Disease Research, Beijing, China Find articles by Xueying Cao 1 , Guangyan Cai Guangyan Cai 1 Department of Nephrology, First Medical Center of Chinese PLA General Hospital, Nephrology Institute of the Chinese People’s Liberation Army, State Key Laboratory of Kidney Diseases, National Clinical Research Center for Kidney Diseases, Beijing Key Laboratory of Kidney Disease Research, Beijing, China Find articles by Guangyan Cai 1 , Xiangmei Chen Xiangmei Chen 1 Department of Nephrology, First Medical Center of Chinese PLA General Hospital, Nephrology Institute of the Chinese People’s Liberation Army, State Key Laboratory of Kidney Diseases, National Clinical Research Center for Kidney Diseases, Beijing Key Laboratory of Kidney Disease Research, Beijing, China Find articles by Xiangmei Chen 1, ✉ , Haiyi Wang Haiyi Wang 3 Department of Radiology, First Medical Center of Chinese PLA General Hospital, Beijing, China Find articles by Haiyi Wang 3, ✉ , Zheyi Dong Zheyi Dong 1 Department of Nephrology, First Medical Center of Chinese PLA General Hospital, Nephrology Institute of the Chinese People’s Liberation Army, State Key Laboratory of Kidney Diseases, National Clinical Research Center for Kidney Diseases, Beijing Key Laboratory of Kidney Disease Research, Beijing, China Find articles by Zheyi Dong 1, ✉ Author information Article notes Copyright and License information 1 Department of Nephrology, First Medical Center of Chinese PLA General Hospital, Nephrology Institute of the Chinese People’s Liberation Army, State Key Laboratory of Kidney Diseases, National Clinical Research Center for Kidney Diseases, Beijing Key Laboratory of Kidney Disease Research, Beijing, China 2 Medical School of Chinese PLA, Beijing, China 3 Department of Radiology, First Medical Center of Chinese PLA General Hospital, Beijing, China 4 GE Healthcare, MR Research China, Beijing, China ✉ Corresponding author. # Contributed equally. Received 2025 Jul 25; Accepted 2026 Mar 15; Collection date 2026 Dec. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . PMC Copyright notice PMCID: PMC13096298 PMID: 42008071 Abstract Objectives To validate blood oxygen level-dependent MRI (BOLD-MRI) for non-invasive discrimination of diabetic nephropathy (DN) vs non-diabetic renal disease (NDRD) and prediction of end-stage renal disease (ESRD) in diabetic kidney disease (DKD). Materials and methods A prospective cohort of 133 biopsy-proven DKD patients underwent BOLD-MRI. The semi-automated 12-layer concentric-objects method was used to analyze BOLD-MRI variables. Prognostic markers for ESRD were identified using univariate and multivariate Cox regression. Feature importance was used to select key diagnostic variables and establish logistic regression and machine-learning differential diagnosis models. Results Among 133 patients (44 DN, 55 NDRD, 34 combined), 20 (15.5%) progressed to ESRD over a mean of 21.8 months. Higher renal medullary R2* (MR2*) (> 24 1/s) reduced ESRD risk by 52% (HR, 0.48) in DKD. Prognostic models integrating pathological grouping, hemoglobin levels, and cysC levels achieved a c -index of 0.90. For the DN and combined groups, MR2*, glomerular grading, interstitial lesions, interstitial fibrosis, and tubular atrophy were predictive of ESRD, with a c -index of 0.91. For differential diagnosis, the random forest (RF) model achieved an AUC of 0.901, with diabetic retinopathy, diabetes duration, albumin, blood urea nitrogen, MR2*, hypertension, and glycosylated hemoglobin as the most contributing factors. For the combined group classified as DN, the AUC of the RF model was 0.791; when classified as NDRD, the AUC was 0.856. Conclusion MR2* shows potential value as a non-invasive diagnostic and prognostic tool in the assessment of DKD. However, BOLD-MRI remains a promising yet exploratory technique that requires external validation and interventional studies before clinical implementation. Critical relevance statement Blood oxygen level-dependent-MRI-derived renal medullary R2* robustly predicts ESRD risk and distinguishes DN without biopsy, offering an immediately translatable, non-invasive biomarker for the precision management of DKD in routine nephrology practice. Trial registration ClinicalTrials.gov, NCT03865914 . Key Points Blood oxygen level-dependent-MRI medullary R2*(MR2*) > 24 s − 1 halves DKD ESRD risk (HR 0.48). MR2* integrated with clinical variables drives c -index to 0.90 for ESRD prognosis. RF leveraging MR2* and clinical traits attains an AUC of 0.901 for diagnosing DN. Graphical Abstract Keywords: BOLD-MRI, Diabetic nephropathy, Differential diagnosis, Prognostic marker, R2* Introduction Diabetic nephropathy (DN), one of the most severe microvascular complications of diabetes mellitus [ 1 ], is characterized by a rising global prevalence due to the lack of early diagnostic tools and effective interventions [ 2 ]. As a leading cause of chronic and end-stage renal disease (ESRD) [ 3 , 4 ], DN significantly contributes to mortality and disability in patients with diabetes [ 5 – 7 ]. Early non-invasive diagnosis and precise prognostic evaluation are critical for reducing the disease burden and guiding clinical decision-making. Early risk stratification in DN identifies rapid progressors sooner, enabling earlier intensive therapy, specialist referral, tailored follow-up, and trial enrollment. It reduces ESRD, helps patients understand their risk, and drives lifestyle changes and better adherence to RAS antagonists, finerenone, and sodium–glucose cotransporter-2 inhibitors (SGLT2i), while reducing overtreatment and optimizing resource use. However, current diagnostic strategies and prognostic tools for DN are inadequate, as they primarily rely on traditional biomarkers such as urinary protein and estimated glomerular filtration rate (eGFR), which fail to accurately assess early renal injury, disease progression, or long-term outcomes [ 8 , 9 ]. Although renal biopsy remains the gold standard, its invasive nature limits its widespread or repeated application in DN management [ 10 ], underscoring the urgent need for non-invasive biomarkers to improve diagnostic and prognostic accuracy. Advances in functional magnetic resonance imaging, particularly blood oxygen level-dependent MRI (BOLD-MRI), offer promising potential for the non-invasive evaluation of renal pathophysiology [ 11 , 12 ]. BOLD-MRI can indirectly assess tissue oxygenation by detecting magnetic field disturbances caused by paramagnetic deoxyhemoglobin [ 13 ], providing insights into renal hypoxia, a key driver of diabetic kidney disease (DKD) progression under the “chronic hypoxia hypothesis” [ 14 ]. The relaxation rate (R2*), which is inversely correlated with tissue oxyhaemoglobin levels [ 15 ], serves as a biomarker for renal oxygenation. Elevated cortical R2* (CR2*) reflects reduced oxygenation, a hallmark of chronic kidney disease (CKD) progression [ 16 ]. Notably, renal R2* negatively correlates with eGFR, whereas reduced medullary R2* (MR2*) is associated with accelerated annual eGFR decline in moderate CKD [ 17 ]. Importantly, MR2* demonstrates sensitivity in detecting medullary hypoxia even in early-stage DN [ 18 ] and correlates with histopathological features such as interstitial fibrosis and tubular atrophy (IFTA) [ 19 , 20 ]. Despite these advances, evidence supporting the role of BOLD-MRI in longitudinal prognosis and histopathological correlation in biopsy-confirmed DN cohorts remains limited. This study aimed to validate BOLD-MRI for non-invasive discrimination of DN vs non-diabetic renal disease (NDRD) and prediction of ESRD in DKD. Materials and methods Study participants This single-center prospective cohort study was conducted from April 2018 to September 2023 at the Department of Nephrology, First Medical Center of the Chinese PLA General Hospital. A total of 140 patients with type 2 diabetes mellitus (T2DM) and kidney disease who underwent renal biopsy and BOLD-MRI examination were enrolled. All adult patients with biopsy results were eligible for BOLD-MRI examination, which was performed within 1 week after biopsy. Inclusion criteria were: (1) age 18–70 years; (2) diagnosis of T2DM; (3) renal impairment, albumin (Alb) excretion > 30 mg/day or GFR < 60 mL/min/1.73 m 2 , or both, of > 3 months’ duration; (4) renal biopsy and BOLD-MRI performed during the same hospitalization; and (5) provision of informed consent. The exclusion criteria were as follows: (1) renal tumors, large renal cysts (maximum diameter ≥ 30 mm), abnormal renal location, congenital renal dysplasia, or renal atrophy; (2) inadequate image quality for analysis due to severe respiratory motion artifacts; (3) significant abdominal edema or exudation; and (4) conditions affecting renal perfusion (e.g., renal artery stenosis or heart failure). This study was approved by the Research Ethics Committee of the Chinese People’s Liberation Army General Hospital (No.S2017-133-01), and informed consent was obtained from all participants. Follow-up was conducted in July 2024 to confirm renal outcomes; the renal endpoints were ESRD or all-cause death. A flowchart of the study is shown in Fig. 1 . Fig. 1. Open in a new tab Flowchart showing the procedure for the selection of study participants. DN, diabetic nephropathy; NDRD, non-diabetic renal disease Renal biopsy and pathological evaluation scoring All patients provided written informed consent before undergoing renal biopsy. Biopsies were performed by an experienced physician, and all specimens were independently reviewed by two pathologists. Any discrepancies were resolved through discussion until a consensus was reached. Tissue was processed and allocated for immunofluorescence, light microscopy, and electron microscopy. DN was diagnosed based on characteristic histological features, including glomerular hypertrophy, thickening of glomerular capillary basement membranes, diffuse mesangial expansion, nodular mesangial sclerosis (Kimmelstiel–Wilson nodules), exudative lesions such as capsular drop or fibrin cap, mesangiolysis, capillary microaneurysms, and hyalinosis of afferent and efferent arterioles [ 17 ]. NDRD was diagnosed based on classical criteria [ 18 ]. Pathological findings were evaluated according to the Pathologic Classification of DN [ 21 ]. Patients with DN were scored using the Renal Pathology Society (RPS) classification to assess the severity of glomerular classification (GC), interstitial lesions (IL; including IFTA and interstitial inflammation [II]), and vascular lesions (VL; including arteriolar hyalinosis [AH] and arteriosclerosis [AS; scored on the worst artery]) [ 21 ] (Fig. 2 ). Fig. 2. Open in a new tab Comparison of MRI and pathological feature images. a – f Are from a 41-year old man with DN; g – l are from a 59-year-old woman with focal segmental glomerulosclerosis; m – r are from a 42-year-old man with DN combined with IgAN.The outer border (red) and inner border (blue) of renal tissue were defined by the operator on the R2* map for TLCO analysis. The color map of the 12 layers computed by the algorithm. DN:LM: the glomeruli exhibit focal, segmental, and mild mesangial cell proliferation, along with diffuse, global, and moderate mesangial matrix expansion. Focal, segmental, and mild endothelial cell proliferation is observed. K-W nodular lesions and exudative changes are present. Some capillary loops appear shrunken and poorly patent. The Bowman’s capsule wall shows diffuse thickening. EM: Significant thickening of glomerular basement membrane (GBM) (900–1097 nm). NDRD: LM: The glomeruli show no significant mesangial cell proliferation or mesangial matrix expansion. Endothelial cell proliferation is not evident, and capillary loops are patent. Parietal epithelial cells show no proliferation, and the Bowman’s capsule wall exhibits segmental thickening. EM: GBM shows no significant thickening (302–307 nm), and podocyte foot processes exhibit segmental effacement. Combined: LM: the glomeruli display focal, segmental, and mild mesangial cell proliferation, along with diffuse, segmental, and moderate mesangial matrix widening. Focal, segmental, and mild endothelial cell proliferation is observed. Mesangiolysis, capillary loop microaneurysms, and exudative changes are present. Some capillary loops are poorly patent, and the Bowman’s capsule wall shows segmental thickening and layering. EM: mesangial cell proliferation and mesangial matrix expansion are evident, with electron-dense deposits in the mesangial region. Capillary loops are poorly patent. Endothelial cell proliferation and segmental widening of the subendothelial space are observed. GBM is significantly thickened (672–804 nm), and podocytes exhibit extensive foot process effacement. LM, light microscopy; EM, electron microscopy Clinical and laboratory indexes Demographic and clinical characteristics of the patients were recorded, including sex, age, diabetes duration, presence of diabetic retinopathy (DR), body mass index (BMI), presence of hypertension, systolic blood pressure (SBP), and diastolic blood pressure. Blood pressure was calculated as the mean of measurements obtained in the fasting and resting states on the day of the BOLD-MRI examination. Laboratory parameters, including fasting blood glucose, glycated hemoglobin (HbA1c), serum creatinine (Scr), blood urea nitrogen (BUN), serum uric acid, cystatin C (Cys-C), hemoglobin (Hb), hematocrit, Alb, triglycerides (TG), total cholesterol (TC), low-density- lipoprotein, proteinuria, 24-h urinary protein (24hUPro), and the presence or absence of glomerular hematuria, were measured within one week prior to MRI scanning. The eGFR was calculated using the 2009 CKD Epidemiology Collaboration (CKD-EPI) creatinine equation, which was also used to stage CKD [ 19 ]. The latest Scr levels were recorded during follow-up to assess renal outcomes. MRI image acquisition and analysis All participants were required to fast for at least 4 h and undergo rigorous breathing training before MRI scanning. Imaging was performed with an eight-channel body array coil (3.0-T MRI Discovery 750; GE Medical Systems). A multiple echo FSPGR sequence (8 echoes ranging from 2.5 to 30 ms, field of view 300 × 300 mm, repetition time 100 ms, bandwidth 31.25 Hz/pixel, flip angle 35°, matrix 192 × 160, number of slices 9, slice thickness 7 mm) was used to acquire transverse relaxation (T2*)-weighted images in the oblique-coronal plane during breath-hold at end-expiration. T2* maps with the corresponding R2* (= 1/T2*) were calculated by fitting the pixel intensity at different echo times to an exponentially decaying function. An internally modified version of the public program ( https://github.com/bmilani/conObj ) was used to generate the R2* images. The “12-layer concentric objects” technique was applied for BOLD-MRI analysis, segmenting the renal parenchyma into 12 equally thick layers extending from the cortex to the medulla and calculating the average R2* value for each layer within the kidney [ 22 ]. The outer and inner boundaries of the renal parenchyma were manually drawn (by a radiologist with approximately 3 years of experience) on the slice, offering the best visualization of the entire renal parenchyma from the R2* map using custom Matlab R2023a (Mathworks) (Fig. 2 ). To assess inter-observer reliability, 50% of the cases were randomly selected and independently delineated by a second radiologist with approximately 5 years of experience. Statistical analysis All statistical analyses were performed using R software (version 4.4.0; R Foundation for Statistical Computing). Categorical variables were expressed as frequencies and proportions, while continuous variables were described as means and standard deviations for normally distributed data and medians with interquartile ranges for non-normally distributed data. Differences between groups were assessed using the chi-square, t -test, or Mann–Whitney U -test, as appropriate. Spearman’s rank correlation coefficient was used to evaluate correlations between variables. Missing data were handled with multiple imputations using the mice package (version 3.16.0). For survival analysis, two datasets were used: one with complete outcome data and another excluding patients with NDRD to include additional pathological scores. Prognostic Cox models were constructed using stepwise regression based on the Akaike Information Criterion and validated through ten-fold cross-validation, with the c -index reported for predictive performance. Cox regression was conducted using the survival package (version 3.7-0). Three models were developed for DN diagnosis: logistic regression, support vector machine (SVM), and random forest (RF), with variable selection based on the importance of the RF algorithm. Model performance was evaluated using the receiver operating characteristic (ROC) curve and the area under the curve (AUC), with a higher AUC indicating better diagnostic efficacy. Diagnostic modeling was performed using R packages e1071 (version 1.7-14), RF (version 4.7-1.1), and cvAUC (version 1.1.4). Results Clinical characteristics Patient demographics, clinical features, and pathology findings According to the research flow chart (Fig. 1 ), 133 patients with T2DM who underwent renal biopsy were included in this study for clinicopathological characterization and BOLD-MRI parameter analysis. In the study cohort, 74.4% ( n = 99) were men, with a mean age of 51.99 ± 10.0 years. The median duration of diabetes was 96.97 months. Regarding complications, 32.3% ( n = 43) had DR, 70.5% ( n = 93) had hypertension, and 43.6% ( n = 58) had massive proteinuria (urinary protein > 3.5 g/24 h). Renal function assessment showed that 79.7% ( n = 106) of patients were in stages 1–3 of CKD, with a baseline eGFR of 62.68 ± 18.35 mL/min/1.73 m 2 . For the BOLD-MRI parameters, the median CR2* was 19.56/s (IQR, 18.08–21.77), and the mean medullary R2* was 23.88 ± 3.88/s. For inter-observer reproducibility, excellent agreement was achieved for CR2* (ICC, 0.965; 95% CI: 0.943–0.978) and medullary R2* (ICC, 0.978; 95% CI: 0.964–0.986) (Supplement Table 1 ). The pathological classification showed that there were 44 cases (33.1%) in the DN group, 55 cases (41.4%) in the NDRD group, and 34 cases (25.6%) in the combined group. There were significant differences among the three groups in terms of parameters such as the course of diabetes, incidence of retinopathy, HbA1c, degree of hematuria, renal function indicators (blood creatinine, urea nitrogen, eGFR), Cys-C, TC, Hb, electrolyte (phosphorus, sodium, potassium), and renal MR2* value ( p < 0.05) (Table 1 and Fig. 3 ). Table 1. Clinical characteristics and laboratory data of patients Parameter Overall DN NDRD DN combine NDRD p value N 133 44 55 34 Sex (male), N (%) 99 (74.4) 33 (75.0) 38 (69.1) 28 (82.4) 0.377 Age, year, mean ± SD 51.99 ± 10.00 51.11 ± 9.76 52.62 ± 10.39 52.12 ± 9.89 0.758 BMI, kg/m 2 , mean ± SD 26.43 ± 3.39 26.32 ± 3.37 26.61 ± 3.83 26.27 ± 2.69 0.878 SBP, mmHg, mean ± SD 136.60 ± 21.65 140.11 ± 24.55 136.60 ± 20.33 132.06 ± 19.37 0.27 DBP, mmHg, median [IQR] 82.00 [70.00, 91.00] 83.00 [71.50, 93.00] 81.00 [71.50, 91.50] 82.00 [68.50, 91.00] 0.89 Diabetes duration, months, median [IQR] 96.97 [25.17, 158.60] 123.32 [75.67, 240.18] 25.23 [4.00, 98.50] 111.40 [63.00, 144.43] < 0.001 DR, N (%) 43 (32.3) 29 (65.9) 3 (5.5) 11 (32.4) < 0.001 hypertension, N (%) 93 (70.5) 36 (81.8) 33 (61.1) 24 (70.6) 0.082 Drinking, N (%) 52 (40.0) 17 (40.5) 18 (33.3) 17 (50.0) 0.298 Smoking, N (%) 59 (45.4) 22 (52.4) 24 (44.4) 13 (38.2) 0.461 HbA1c, %, median [IQR] 6.80 [6.20, 7.95] 7.85 [6.58, 8.88] 6.55 [6.20, 7.20] 6.50 [6.00, 7.60] 0.001 HbA1c ≥ 7, %, N (%) 55 (44.7) 30 (71.4) 14 (29.2) 11 (33.3) < 0.001 FBG, mmol/L, median [IQR] 6.15 [5.18, 7.43] 6.58 [5.20, 8.79] 6.15 [5.34, 7.19] 5.78 [4.61, 6.86] 0.072 Scr, umol/L, median [IQR] 103.80 [80.50, 163.80] 130.45 [100.50, 180.38] 93.30 [71.25, 152.30] 94.25 [76.82, 129.38] 0.005 eGFR, mL/min/1.73 m 2 , median [IQR] 62.68 [33.81, 90.60] 53.26 [30.52, 68.88] 68.96 [46.24, 97.89] 74.25 [48.91, 93.46] 0.01 BUN, mmol/L, median [IQR] 7.48 [5.55, 10.45] 8.80 [7.40, 12.39] 6.14 [4.48, 10.23] 6.92 [5.56, 9.35] 0.001 Ua, mol/L, mean ± SD 376.91 ± 97.40 375.62 ± 82.91 379.41 ± 120.54 374.55 ± 72.04 0.971 Cys-C, mg/L, median [IQR] 1.23 [1.00, 1.73] 1.46 [1.21, 2.28] 1.13 [0.85, 1.41] 1.18 [1.02, 1.58] 0.003 CKD stage, N (%) 0.037 1 34 (25.6) 6 (13.6) 19 (34.5) 9 (26.5) 2 37 (27.8) 9 (20.5) 16 (29.1) 12 (35.3) 3 35 (26.3) 18 (40.9) 7 (12.7) 10 (29.4) 4 18 (13.5) 7 (15.9) 9 (16.4) 2 (5.9) 5 9 (6.8) 4 (9.1) 4 (7.3) 1 (2.9) Gh, N (%) 56 (42.1) 21 (47.7) 28 (50.9) 7 (20.6) 0.012 24hUPro, g/24 h, median [IQR] 2.93 [1.24, 5.37] 3.33 [1.58, 4.94] 2.38 [1.23, 5.78] 2.93 [1.15, 5.94] 0.902 TC, mmol/L, median [IQR] 4.77 [3.88, 6.00] 4.32 [3.71, 5.19] 5.05 [4.19, 6.31] 5.07 [3.67, 6.22] 0.046 TG, mmol/L, median [IQR] 2.08 [1.61, 2.84] 1.94 [1.61, 2.58] 2.35 [1.68, 3.11] 1.98 [1.42, 2.82] 0.265 LDL, mmol/L, median [IQR] 2.95 [2.18, 4.01] 2.68 [2.02, 3.62] 3.09 [2.44, 4.11] 3.28 [2.14, 4.36] 0.079 HDL, mmol/L, median [IQR] 1.00 [0.87, 1.14] 0.97 [0.90, 1.10] 1.03 [0.85, 1.16] 1.00 [0.87, 1.15] 0.957 TP, g/L, mean ± SD 57.69 ± 11.29 59.74 ± 8.42 56.68 ± 12.59 56.66 ± 12.22 0.255 ALB, g/L, median [IQR] 34.30 [25.80, 40.40] 35.90 [31.60, 38.73] 33.60 [23.95, 41.15] 34.55 [23.13, 41.20] 0.583 TIBC, mol/L, mean ± SD 41.38 ± 9.83 41.96 ± 8.73 41.72 ± 11.50 40.14 ± 8.55 0.722 SI, µmol/L, median [IQR] 13.20 [9.25, 16.75] 9.90 [8.50, 14.10] 15.20 [10.60, 18.00] 14.50 [12.00, 17.00] 0.019 SF, g/L, median [IQR] 273.25 [114.15, 440.20] 328.15 [236.20, 481.12] 232.80 [116.10, 427.80] 119.50 [77.15, 418.00] 0.06 RBC, g/L, mean ± SD 4.22 ± 0.71 3.94 ± 0.70 4.37 ± 0.68 4.31 ± 0.71 0.008 Hb, g/L, mean ± SD 126.23 ± 22.18 116.45 ± 21.44 130.78 ± 21.13 131.53 ± 21.16 0.002 MCH, pg, mean ± SD 30.05 ± 1.60 29.63 ± 1.36 30.05 ± 1.82 30.60 ± 1.39 0.012 MCHC, g/L, median [IQR] 344.50 [336.00, 351.25] 344.50 [336.00, 351.00] 343.00 [334.00, 352.50] 346.00 [338.00, 352.00] 0.665 HCT, %, median [IQR] 0.35 [0.23, 0.40] 0.34 [0.29, 0.37] 0.35 [0.13, 0.39] 0.38 [0.26, 0.42] 0.302 MCV, fL, median [IQR] 86.35 [84.38, 89.53] 85.80 [84.00, 89.00] 86.20 [84.35, 89.65] 87.90 [85.20, 89.80] 0.327 Hp, mg/dL, (median [IQR]) 130.00 [71.80, 178.50] 149.00 [116.00, 193.00] 131.50 [89.65, 172.75] 70.10 [29.02, 128.50] < 0.001 Na, mmol/L median [IQR] 141.10 [139.50, 142.60] 141.30 [140.07, 142.72] 140.10 [138.38, 141.62] 141.40 [139.90, 142.80] 0.022 K, umol/L, median [IQR] 3.98 [3.68, 4.32] 4.14 [3.84, 4.47] 3.82 [3.66, 4.17] 3.87 [3.67, 4.22] 0.035 P, mmol/L, median [IQR] 1.28 [1.12, 1.38] 1.32 [1.23, 1.52] 1.23 [1.08, 1.34] 1.28 [1.07, 1.38] 0.021 cortex R2*, 1/s, median [IQR] 19.56 [18.08, 21.77] 19.53 [18.08, 20.48] 19.68 [18.23, 22.62] 19.41 [17.88, 21.61] 0.25 medulla R2*, 1/s, mean ± SD 23.88 ± 3.88 23.00 ± 3.63 24.77 ± 3.65 23.56 ± 4.32 0.01 Follow-up time, months, median [IQR] 15.00 [8.00, 30.00] 15.00 [8.00, 25.50] 14.00 [7.50, 29.50] 24.00 [8.25, 53.75] 0.318 Latest Scr, umol/L, median [IQR] 113.20 [82.10, 170.00] 139.30 [99.60, 300.00] 91.50 [70.42, 135.45] 119.35 [90.75, 145.27] 0.002 Latest eGFR, mL/min/1.73 m 2 , median [IQR] 62.92 [35.00, 88.89] 44.54 [19.13, 73.84] 78.43 [47.72, 96.23] 59.17 [44.63, 74.90] 0.007 Latest.24hUPro, g/24 h, median [IQR] 1.44 [0.65, 3.71] 2.95 [0.94, 3.71] 1.20 [0.85, 3.01] 1.23 [0.42, 4.50] 0.342 Endpoint, N (%) renal replacement therapy, N (%) 14 (10.5) 12 (27.3) 1 (1.8) 1 (2.9) < 0.001 all cause mortality, N (%) 5 (3.8) 3 (6.8) 0 (0.0) 2 (5.9) < 0.001 entering ESRD, N (%) 20 (15.5) 15 (34.9) 3 (5.6) 2 (6.2) < 0.001 Open in a new tab Data are presented as n (%), mean ± SD, or median (IQR) BMI body mass index, SBP systolic blood pressure, DBP diastolic blood pressure, DR diabetic retinopathy, HbA1c glycated hemoglobin, FBG fasting blood glucose, Scr serum creatinine, eGFR estimated glomerular filtration rate, BUN blood urea nitrogen, Ua blood uric acid, Cys-C serum cystatin C, CKD chronic kidney disease, Gh glomerular hematuria, 24hUPro 24-h urinary protein, TC total cholesterol, TG triglycerides, LDL low-density lipoprotein, TP total serum protein, ALB serum albumin, TIBC total iron binding capacity, SI serum iron, SF serum ferritin, RBC red blood cells, Hb hemoglobin, MCH mean hemoglobin content of red blood cell, MCHC mean corpuscular hemoglobin concentration, HCT hematocrit, MCV mean red blood cell volume, Hp haptoglobin, Na sodium, K potassium, P phosphorus Fig. 3. Open in a new tab Violin plot of BOLD parameters with different pathological and renal outcomes. A , B Pathological group. C , D Renal outcome: entering ESRD or not. a – n Renal Pathological Score (RPS)–based stratification: Glomerular classification, Interstitial lesions, Interstitial fibrosis and atrophy (IFTA), Interstitial inflammation, Vascular lesions, Arteriolar hyalinosis, and Arteriosclerosis The median follow-up time was 15 months, with a mean follow-up of 21.8 months. At the last follow-up, the mean Scr level was 113.20 μmol/L, corresponding to a mean eGFR of 62.92 mL/min/1.73 m 2 . Endpoint analysis showed that 14 cases (10.5%) received renal replacement therapy, 5 cases (3.8%) experienced all-cause mortality, and 20 cases (15.5%) progressed to ESRD. There was a significant difference in the incidence of endpoint events among the three groups (DN group > combined group > NDRD group, p < 0.001) (Table 1 ). Further analysis of the clinicopathological characteristics of ESRD progressors showed that this subgroup had a longer duration of diabetes ( p = 0.017), lower baseline eGFR ( p < 0.001), more severe proteinuria ( p = 0.01), more severe anemia ( p < 0.001), and a lower renal MR2* value (20.48 [19.18–23.15] vs 24.31 [21.59–26.77], p < 0.001) (Table 2 ). Table 2. Comparison of clinicopathological characteristics of patients entering ESRD or non-ESRD Parameter Overall Non- ESRD Entering ESRD p N 129 109 20 Sex (male), N (%) 97 (75.2) 85 (78.0) 12 (60.0) 0.153 Age, year, mean ± SD 53.00 [45.00, 60.00] 53.00 [45.00, 60.00] 52.50 [44.75, 58.50] 0.81 BMI, kg/m 2 , mean ± SD 26.55 (3.36) 26.73 (3.40) 25.55 (2.97) 0.122 SBP, mmHg, mean ± SD 136.64 (21.92) 135.06 (21.46) 145.25 (22.93) 0.077 DBP, mmHg, median [IQR] 82.00 [70.00, 91.00] 81.00 [70.00, 91.00] 86.00 [76.75, 89.75] 0.283 Diabetes duration, months, median [IQR] 97.00 [24.97, 158.77] 88.43 [24.00, 151.77] 122.20 [91.40, 204.61] 0.017 DR, N (%) 42 (32.6) 31 (28.4) 11 (55.0) 0.038 hypertension, N (%) 90 (70.3) 74 (68.5) 16 (80.0) 0.444 Pathological group (%) < 0.001 DN 43 (33.3) 28 (25.7) 15 (75.0) NDRD 54 (41.9) 51 (46.8) 3 (15.0) DN combine NDRD 32 (24.8) 30 (27.5) 2 (10.0) HbA1c, %, median [IQR] 6.80 [6.20, 7.90] 6.80 [6.20, 7.90] 6.75 [6.05, 8.07] 0.952 FBG, mmol/L, median [IQR] 6.13 [5.18, 7.43] 6.22 [5.27, 7.32] 5.42 [4.74, 7.71] 0.267 Scr, µmol/L, median [IQR] 103.80 [76.70, 168.10] 97.60 [75.40, 135.40] 183.30 [154.18, 304.03] < 0.001 eGFR, mL/min/1.73 m 2 , median [IQR] 62.68 [33.81, 90.60] 68.47 [49.21, 95.86] 27.30 [18.92, 41.06] < 0.001 BUN, mmol/L, median [IQR] 7.43 [5.52, 10.35] 6.99 [5.03, 9.47] 9.97 [8.82, 13.64] < 0.001 Ua, mol/L, mean ± SD 375.99 (97.97) 381.26 (99.77) 347.29 (83.95) 0.117 Cys-C, mg/L, median [IQR] 1.23 [0.99, 1.68] 1.17 [0.94, 1.49] 2.27 [1.57, 2.83] < 0.001 24 hUPro, g/24 h, median [IQR] 2.93 [1.24, 5.37] 2.38 [1.09, 5.10] 4.63 [3.21, 6.60] 0.01 TC, mmol/L, median [IQR] 4.76 [3.87, 6.00] 4.76 [3.87, 6.02] 4.90 [3.96, 5.90] 0.816 TG, mmol/L, median [IQR] 2.08 [1.58, 2.87] 2.10 [1.48, 2.90] 1.93 [1.68, 2.77] 0.961 ALB, g/L, median [IQR] 34.40 [26.60, 40.40] 35.80 [26.20, 41.20] 33.25 [27.52, 36.30] 0.149 SI, µmol/L, median [IQR] 13.20 [9.25, 16.75] 14.85 [11.12, 17.85] 8.60 [6.70, 10.70] < 0.001 SF, g/L, median [IQR] 272.70 [114.80, 446.00] 242.90 [113.50, 427.80] 349.75 [152.20, 589.45] 0.137 Hb, g/L, mean ± SD 126.46 (22.28) 130.07 (20.88) 106.75 (19.54) < 0.001 MCH, pg, mean ± SD 30.06 (1.61) 30.20 (1.66) 29.22 (0.96) 0.001 HCT, %, median [IQR] 0.44 [0.37, 13.00] 0.47 [0.38, 13.00] 0.34 [0.31, 10.19] 0.021 Hp, mg/dL, (median [IQR]) 130.00 [72.40, 178.25] 129.00 [71.80, 166.50] 172.00 [96.90, 217.00] 0.139 Na, mmol/L, median [IQR] 141.05 [139.50, 142.60] 141.10 [139.15, 142.55] 141.00 [140.25, 143.30] 0.28 K, µmol/L, median [IQR] 4.01 [3.68, 4.33] 3.93 [3.68, 4.24] 4.36 [3.80, 4.62] 0.065 P, mmol/L, median [IQR] 1.27 [1.12, 1.38] 1.26 [1.09, 1.37] 1.31 [1.22, 1.44] 0.21 cortex R2*, 1/s, median [IQR] 19.55 [17.91, 21.67] 19.67 [18.27, 21.77] 18.64 [16.51, 20.21] 0.082 medulla R2*, 1/s, mean ± SD 23.96 [21.06, 26.67] 24.31 [21.59, 26.77] 20.48 [19.18, 23.15] < 0.001 Open in a new tab 24hUPro 24-h urinary protein, ALB albumin, BMI body mass index, BUN blood urea nitrogen, Cys-C cystatin C, DBP diastolic blood pressure, DN diabetic nephropathy, DR diabetic retinopathy, eGFR estimated glomerular filtration rate, FBG fasting blood glucose, Hb hemoglobin, HbA1c glycated hemoglobin, HCT hematocrit, Hp haptoglobin, K potassium, MCH mean corpuscular hemoglobin, Na sodium, NDRD non-diabetic renal disease, P phosphorus, SBP systolic blood pressure, Scr serum creatinine, SF serum ferritin, SI serum iron, TC total cholesterol, TG triglycerides, Ua blood uric acid Correlation analysis of clinicopathological features Violin Plot Analysis (Fig. 3 ): analysis based on violin plots revealed that renal MR2* values revealed significant differences among ESRD progression status ( p < 0.001) and pathological classification (DN group vs NDRD group vs combined group, p = 0.01). In contrast, the renal CR2* values did not show statistical significance among the groups ( p > 0.05). Spearman correlation analysis: to determine whether BOLD-MRI parameters independently predict renal prognosis from conventional clinical indicators, we conducted a Spearman correlation analysis on the clinical characteristics and R2* parameters of 133 patients. The results indicated that renal CR2* had no significant correlation with any clinical indicators (including eGFR, Scr, and proteinuria). Conversely, renal MR2* demonstrated a weak positive correlation with eGFR ( r = 0.20, p = 0.02), a weak negative correlation with Cys-C ( r = −0.28, p = 0.003), and a moderate positive correlation with Hb levels ( r = 0.34, p < 0.001) (Table 3 and Fig. 4A ). Table 3. Correlation analysis between BOLD-R2* and clinicopathological parameters Parameter Cortex R2* Medulla R2* r p value r p value Male 0.1 0.28 0.01 0.93 Hypertension −0.1 0.26 −0.13 0.93 BMI 0.01 0.88 0.07 0.13 Diabetes duration 0.02 0.86 −0.12 0.4 DR −0.13 0.14 −0.23 0.18 SBP −0.14 0.11 −0.17 0.009 24hUP −0.09 0.33 −0.18 0.22 Gh −0.12 0.18 −0.11 0.04 FBG −0.08 0.37 −0.02 0.21 Alb 0.09 0.33 0.2 0.03 HbA1c 0.01 0.94 −0.01 0.93 Scr 0.11 0.21 −0.21 0.02 eGFR −0.09 0.32 0.2 0.02 Cys-C 0.05 0.59 −0.28 0.003 TC −0.05 0.53 −0.13 0.14 LDL −0.07 0.44 −0.11 0.2 Hb 0.13 0.15 0.34 < 0.001 HCT 0.19 0.04 0.15 0.1 GC −0.01 0.92 −0.28 0.01 IL 0.11 0.35 −0.17 0.13 IFTA 0.08 0.49 −0.18 0.12 II 0.02 0.89 −0.05 0.67 VL −0.03 0.81 −0.02 0.87 AH −0.06 0.58 −0.14 0.21 AS −0.04 0.73 0.02 0.84 Open in a new tab 24hUP 24-h urinary protein, AH arteriolar hyalinosis, Alb albumin, AS arteriosclerosis, BMI body mass index, Cys-C cystatin C, DR diabetic retinopathy, eGFR estimated glomerular filtration rate, FBG fasting blood glucose, GC Glomerular classification, Gh glomerular hematuria, Hb hemoglobin, HbA1c glycated hemoglobin, HCT hematocrit, IFTA interstitial fibrosis and tubular atrophy, IL interstitial lesions, II interstitial inflammation, LDL low-density lipoprotein, SBP systolic blood pressure, Scr serum creatinine, TC total cholesterol, VL Vascular lesions Fig. 4. Open in a new tab Correlation heatmap between BOLD parameters and Clinicopathological features. A Clinical parameter correlations in the DKD cohort ( n = 133). B Clinicopathological analysis in non-Pure NDRD subcohort ( n = 78). AH, arteriolar hyalinosis; AS, arteriosclerosis; GC, glomerular classification; IL, interstitial lesions; IFTA, interstitial fibrosis and tubular atrophy; II, interstitial inflammation; VL, vascular lesions Further analysis in patients with non-pure NDRD: further analysis was conducted on the renal pathological RPS scores and imaging parameters (CR2* and MR2* values) of 78 patients with non-pure NDRD (DN and combined groups). The results showed weak correlations between renal CR2* and MR2* values and the seven pathological indicators. Specifically, MR2* values exhibited a negative correlation with glomerular grading ( r = −0.28, p < 0.01). Notably, the degree of pathological tissue damage (including glomerular grading, IL, and IFTA) showed moderate correlations with clinical renal function indicators (eGFR and Cys-C) and the degree of anemia (Table 3 and Fig. 4B ). ESRD prognostic evaluation model Prognostic model for the entire population cohort Using Cox proportional hazards regression, we constructed an ESRD prognostic model for 129 patients with DKD (4 lost to follow-up). Kaplan–Meier univariate survival analysis showed that the survival curve of the DN group consistently lay below those of the NDRD and combined groups throughout the study period (Fig. 5a ). The log-rank test indicated significant differences in survival times among the three groups ( p < 0.001), with the DN group exhibiting the worst prognosis and a median survival of 40 months. Univariate Cox regression analysis revealed that patients with DN had a 6.48-fold higher risk of progressing to ESRD than those with NDRD (Fig. 5A ). The survival curve of the higher MR2* group remained above that of the lower group during the study period, with significant differences in survival times ( p < 0.001) (Fig. 5c ). In contrast, CR2* did not provide prognostic information (Fig. 5b ). Additionally, clinical parameters such as BMI, disease duration, SBP, renal function (eGFR, Scr, BUN, Cys-C), and Hb level were predictive of prognosis ( p < 0.05) (Fig. 5A ). Multivariate Cox proportional hazards regression analysis identified pathological grouping, Hb level, and Cys-C level as independent prognostic factors for renal survival in patients with DKD (Fig. 5B ). Pathological grouping and Cys-C maintained statistical significance in the multivariate analysis, with a c -index of 0.90 from ten-fold cross-validation. Fig. 5. Open in a new tab Univariate and multivariate Cox regression and KM Survival analysis of ESRD risk. A – D Univariate and multivariate Cox regression, A forest plot: univariate Cox regression analysis for the entire population, B forest plot: multivariate Cox regression analysis for the entire population, C Forest plot: univariate Cox regression analysis for the RPS assessment population, and D forest plot: multivariate Cox regression analysis for the RPS assessment population. a – l KM Survival analysis, a – c full data population, d – I RPS assessment population Prognostic model for DN and combined groups (RPS assessment population) For the 75 patients in the DN and combined groups (with 3 lost to follow-up), we quantified pathological RPS scores and incorporated clinical and pathological parameters to construct a Cox proportional hazards regression model for ESRD progression in patients with non-pure NDRD. Univariate survival analysis demonstrated that MR2*, glomerular grading, IL, and IFTA were predictive of ESRD in pathologically diagnosed patients with DN, whereas other RPS scores lacked prognostic value (Fig. 5d–k ). For every one-unit increase in glomerular grading, IL, and IFTA scores, the risk of ESRD increased by 2.79, 2.77, and 6.29 times, respectively (Fig. 5C ). Multivariate Cox proportional hazards regression analysis indicated that 24hUPro, eGFR, Hb, MR2*, glomerular grading, IL, and IFTA were independent prognostic factors for renal survival in patients with DKD (Fig. 5D ). The prognostic model had a c -index of 0.91 from ten-fold cross-validation. Differential diagnosis model Comparison between DN and NDRD groups Compared with patients in the NDRD group, those in the DN group had a longer duration of diabetes, a higher incidence of DR, higher HbA1c levels, better-controlled lipid profiles, lower estimated eGFR, and lower Hb levels. The renal MR2* was significantly lower in the DN group than in the NDRD group (23.00 ± 3.63 vs 24.77 ± 3.65, p = 0.018). Univariate/multivariate analysis To evaluate the independent contribution of Medulla R2* to differential diagnosis, we conducted univariate and multivariate logistic regression analyses (Supplementary Fig. 1 ). In univariate analysis (Panel A), Medulla R2* demonstrated a significant association with DKD status (OR 0.87, 95% CI: 0.77–0.98, p = 0.022), indicating its standalone discriminatory capability. However, in multivariate analysis adjusting for established clinical risk factors including DR, diabetes duration, HbA1c, hypertension, ALB, and BUN (Panel B), the effect estimate attenuated to non-significance (OR 0.94, 95% CI: 0.77–1.15, p = 0.571). Two-stage diagnostic performance of MR2* in differential diagnosis To evaluate both the standalone and incremental diagnostic value of MR2*, we employed a two-stage modeling approach across three cohorts (Fig. 6 ). Feature importance of the RF model was used to identify parameters that could distinguish between DN and NDRD. Seven parameters were selected: DR, diabetes duration, Alb, BUN, renal MR2*, hypertension, and HbA1c (Fig. 6A ). Fig. 6. Open in a new tab Differential diagnosis model and feature importance ranking based on BOLD-R2*. A Feature importance ranking. B Differential diagnosis cohort 1: DN Vs NDRD. C Differential diagnosis cohort 2: DN combine NDRD is classified into DN. D Differential diagnosis cohort 3: DN combine NDRD is classified into NDRD Independent performance of MR2*: when used as the sole predictor, Medulla R2* demonstrated modest but consistent discriminatory capability across all three cohorts. The AUC values for MR2* alone were 0.64 in Cohort 1 (Fig. 6B ), 0.617 in Cohort 2 (Fig. 6C ), and 0.610 in Cohort 3 (Fig. 6D ). These findings indicate that MR2* possesses intrinsic diagnostic information for distinguishing DKD, although its standalone performance is limited compared to comprehensive clinical assessment. The consistency of AUC values across diverse patient populations (range 0.610–0.641) supports the robustness and generalizability of this imaging biomarker. Performance of integrated models: when MR2* was integrated with clinical variables using machine learning algorithms, diagnostic performance improved substantially across all cohorts. Diagnostic models were constructed using logistic regression (LR), SVM, and RF, and validated via ten-fold cross-validation. The AUC values for the models were RF (0.901), LR (0.876), and SVM (0.845). The RF model demonstrated the highest diagnostic efficacy, with an accuracy of 0.857, a sensitivity of 0.801, a specificity of 0.901, and an F1-score of 0.832 (Table 4 and Fig. 6B ). When the combined group was classified as DN (Cohort 2), the AUC values were RF (0.791), LR (0.750), and SVM (0.764), with RF demonstrating the highest diagnostic efficacy. When the combined group was classified as NDRD (Cohort 3), the AUC values were RF (0.856), LR (0.786), and SVM (0.761) (Table 4 and Fig. 6C, D ). Table 4. Diagnostic efficacy of the differential diagnostic model between DN and NDRD Model Accuracy Sensitivity Specificity Pos pred value Neg pred value F1 AUC Model 1: DN Vs NDRD RF 0.857 0.801 0.901 0.866 0.850 0.832 0.901 LR 0.831 0.749 0.896 0.853 0.817 0.797 0.876 SVM 0.856 0.807 0.895 0.860 0.853 0.832 0.845 Model 2: DN combine NDRD is classified into DN RF 0.708 0.849 0.595 0.626 0.831 0.721 0.791 LR 0.718 0.780 0.669 0.653 0.792 0.711 0.750 SVM 0.710 0.726 0.698 0.658 0.761 0.690 0.764 Model 3: DN combine NDRD is classified into NDRD RF 0.743 0.889 0.695 0.485 0.951 0.628 0.856 LR 0.777 0.699 0.803 0.533 0.892 0.605 0.786 SVM 0.720 0.788 0.699 0.457 0.911 0.579 0.761 Open in a new tab DN diabetic nephropathy, LR logistic regression, NDRD non-diabetic renal disease, RF random forest, SVM support vector machine Discussion DKD is one of the leading causes of ESRD. Early diagnosis and accurate prognostic assessment are crucial for delaying disease progression. However, traditional clinical indicators, such as eGFR and proteinuria, have certain limitations for early diagnosis and prognostic evaluation [ 23 ]. In recent years, BOLD-MRI, as a non-invasive, contrast agent-free, and radiation-free imaging technique, has gradually demonstrated significant value in the early diagnosis and renal function assessment of CKD [ 24 – 26 ]. Through a prospective cohort design, this study systematically explored, for the first time, the clinicopathological correlations and prognostic diagnostic value of BOLD-MRI in DKD, filling a research gap in this field. Based on a cohort of 133 patients with T2DM confirmed by renal biopsy, this study found a significant association between MR2* and the progression of renal function, as well as pathological diagnosis. Furthermore, the study innovatively developed a multimodal prediction and diagnostic model integrating BOLD-MRI features, clinical indicators, and renal pathological parameters, overcoming the limitations of traditional single-biomarker assessments. This provides a clinically translatable new tool for non-invasive quantitative analysis of renal tissue hypoxia and individualized treatment decision-making. BOLD-MRI exhibits excellent complementarity with traditional indicators, such as eGFR and proteinuria. While eGFR primarily reflects glomerular filtration function, and proteinuria is influenced by multiple factors, both have limited sensitivity in early DKD and fail to reflect renal oxygenation status [ 27 ]. In contrast, BOLD-MRI, by quantifying renal oxygenation status, can sensitively detect changes and progression in renal function. Numerous studies have confirmed that cortical and medullary oxygenation function in patients with CKD is significantly lower than that in healthy controls, with medullary oxygenation decline serving as an effective predictor of renal function progression [ 28 , 29 ]. Feng et al demonstrated that BOLD-MRI can assess early renal hypoxia in DKD, suggesting that MR2* values may be a more sensitive predictor of early DKD [ 18 ]. We found that MR2* was significantly higher than CR2* ( p < 0.001), consistent with its physiological hypoxic characteristics and aligned with the renal oxygen supply–demand balance mechanism [ 20 , 27 ]. Specifically, the high metabolic demand of the renal medulla, primarily due to active tubular reabsorption, and its relatively low blood supply result in a substantial oxygen partial pressure gradient, whereas the cortex maintains better oxygenation due to high blood flow perfusion. Furthermore, the higher MR2* in the NDRD group may be related to the primary pathological features of NDRD (mainly membranous nephropathy, 47.2%, IgA nephropathy, 21.8%, focal segmental glomerulosclerosis, and acute or chronic interstitial damage, 25.4%), among which NDRD cases involving interstitial VL accounted for 71%. Interstitial VL, inflammatory infiltration, and fibrosis exacerbate medullary hypoxia. In contrast, the lower MR2* in the DN group may be attributed to early glomerular hyperfiltration and hyperperfusion, which induce a hypoxic state in the renal medulla during the early stages of CKD (manifested as elevated medullary R2* values). However, as DN progresses, the sustained high metabolic load ultimately leads to an imbalance between oxygen demand and supply and a reversal of the cortical–medullary oxygenation gradient, resulting in improved medullary oxygenation and reduced MR2* [ 30 – 32 ]. In addition, the lower MR2* observed in the ESRD group may reflect reduced tubular workload and oxygen consumption. In early CKD, glomerular hyperfiltration and hyperperfusion induce medullary hypoxia, elevating R2* [ 28 ]. As DN progresses, impaired tubular concentration and sodium reabsorption, due to decreased GFR and altered tubuloglomerular feedback, reduce metabolic demand, thereby alleviating medullary hypoxia and leading to a decline in R2* [ 18 , 33 , 34 ] For example, Feng et al reported that diabetic patients with microalbuminuria had significantly higher MR2* than healthy controls, whereas those with macroalbuminuria showed significantly lower MR2* than the microalbuminuria group. Previous studies on CKD have demonstrated that R2* values are closely associated with renal pathological sclerosis and tubulointerstitial fibrosis, highlighting their importance in evaluating renal histological changes [ 19 , 20 ]. In this cohort, significant clinical heterogeneity was observed among the different pathological groups (DN, NDRD, and combined groups), with the DN group exhibiting the highest incidence of endpoint events ( p < 0.001). This finding aligns with the extensive and complex pathological changes in glomerulosclerosis and interstitial fibrosis characteristic of DN [ 35 ], further corroborating the global epidemiological trend of DN as a leading cause of ESRD. Notably, the subgroup progressing to ESRD demonstrated a longer duration of diabetes, lower baseline eGFR, more severe proteinuria [ 23 ], and lower MR2* values (all p < 0.05), underscoring the critical importance of early identification and intervention in patients who are high-risk. CR2* and MR2* values showed weak correlations with conventional clinical indicators, such as eGFR and proteinuria. Specifically, MR2* exhibited a weak negative correlation with glomerular lesion grading ( r = −0.28, p = 0.01). Medullary hypoxia may exacerbate inflammation and fibrosis through the activation of hypoxia-inducible factor pathways [ 36 ]. Although our study did not find a significant correlation between IFTA and R2*, both parameters independently predicted prognosis. The integration of BOLD-MRI with pathological indicators may optimize prognostic models, offering a novel, non-invasive tool for the precise diagnosis and treatment of DKD. In constructing the ESRD prognostic assessment model, Cox regression analysis of the entire cohort revealed that pathological grouping, Hb, Cys-C, and MR2* were independent predictors of ESRD prognosis (all p < 0.05), consistent with previous reports [ 37 ]. In the DN and combined groups, MR2*, IL, and IFTA were significant in the univariate Cox regression analysis ( p < 0.05); however, their significance decreased in the multivariate analysis ( p > 0.05). This reduction in significance in the multivariate analysis may reflect interactions or confounding effects among variables, which can alter the individual significance of the predictors. Notably, despite the decreased significance of these indicators, the model demonstrated excellent predictive performance, with a c -index of 0.91, suggesting that these indicators may still hold clinical relevance after controlling for confounding factors [ 38 – 40 ]. The integration of multimodal data enables more precise risk stratification, facilitating the identification of high-risk patients and optimizing therapeutic decision-making. We acknowledge that the independent contribution of MR2* is indeed modest compared to DR. As shown in Supplementary Fig. 1 , MR2* demonstrated a significant association in univariate analysis (OR 0.87, p = 0.022), but the effect attenuated in multivariate analysis (OR 0.94, p = 0.571). This attenuation primarily reflects the strong predictive power of DR (OR 31.13, p < 0.001), which captures overlapping microvascular information. Importantly, the effect direction remained consistently protective (OR < 1) with minimal magnitude change, supporting the biological validity of MR2* as a quantitative marker of renal medullary oxygenation status. Similarly, in Fig. 6 , the AUC values for MR2* alone were moderate (0.610–0.641). However, this moderate degree of independent contribution is in line with clinical expectations and does not reduce the research value of BOLD-MRI. Because MR2* is a non-invasive and potentially valuable imaging biomarker, it contributes meaningful research value to the comprehensive assessment of DN. For the diagnostic model, we used feature importance to identify discriminative parameters. The machine learning–based diagnostic model (RF: AUC = 0.901) outperformed traditional logistic regression (AUC = 0.876), highlighting its superiority in handling complex clinical data and feature selection [ 41 ]. This exploratory study on the diagnostic and prognostic value of BOLD-MRI in DKD patients with renal biopsy provides a basis for clinical decision-making. However, considering the cost and clinical benefits of BOLD-MRI, MR2* may be used as a supplementary tool in specific cases (such as when other biomarkers are not applicable or are insufficiently accurate). This study also has several limitations. First, the single-center design may restrict the generalizability and extrapolation of the findings, necessitating further validation through multicenter studies with larger sample sizes. Second, BOLD-MRI signals can be influenced by various physiological factors, including respiratory artifacts, blood pH, temperature, red blood cell volume, hydration status, and salt intake [ 42 ]. To mitigate these potential confounders, several measures were implemented in this study: a fasting period of at least 4 h was required before imaging, images with compromised quality were excluded, and a double-blind approach was adopted for data processing. Additionally, machine learning methods were employed during data analysis and modeling to control for confounding factors. The median follow-up in this study was 15 months, indicating that longer observation will be required to fully define the long-term prognosis. Future research should focus on multicenter longitudinal cohort studies to validate the generalizability of the model, explore the combined application of BOLD-MRI with emerging biomarkers, and develop AI-driven radiomics models to enhance diagnostic sensitivity and specificity. These efforts will contribute to refining the precision medicine framework for DKD. BOLD-MRI provides a novel non-invasive tool for risk stratification and individualized intervention in DKD. Conclusion BOLD-MRI technology, particularly MR2*, shows potential value as a non-invasive diagnostic and prognostic tool in the assessment of DKD. However, BOLD-MRI remains a promising, yet exploratory, technique that requires external validation and interventional studies before clinical implementation. Supplementary information ELECTRONIC SUPPLEMENTARY MATERIAL (248.9KB, pdf) Acknowledgements The authors are grateful to all participants for their time and effort. We thank the patients involved in the study. We thank Editage for editing and proofreading the text. Abbreviations 24hUPro 24-h urine protein AH Arteriolar hyalinosis Alb Albumin AS Arteriosclerosis AUC Area under the curve BMI Body mass index BOLD-MRI Blood–oxygen-level-dependent magnetic resonance imaging BUN Blood urea nitrogen CKD Chronic kidney disease CR2* Cortical R2* Cys-C Cystatin C DKD Diabetic kidney disease DN Diabetic nephropathy DR Diabetic retinopathy eGFR Estimated glomerular filtration rate ESRD End-stage renal disease GC Glomerular classification Hb Hemoglobin HbA1c Glycosylated hemoglobin IFTA Interstitial fibrosis and tubular atrophy II Interstitial inflammation IL Interstitial lesions LDL Low-density lipoprotein MR2* Renal medullary R2* NDRD Non-diabetic renal disease RF Random forest SBP Systolic blood pressure Scr Serum creatinine T2DM Type 2 diabetes mellitus TC Total cholesterol TG Triglycerides VL Vascular lesions Author contributions Qian Wang contributed to manuscript writing, patient enrollment, data collection, and statistical analysis. Shaopeng Zhou and Chaobo Li were responsible for the implementation of BOLD examinations and renal region delineation. Yue Niu and Qiang Lyu contributed to data analysis and model building. Pu Chen assessed renal pathology. Xiaojing Zhang implemented BOLD examinations. Lizhi Xie provided technical support and image processing. Wanjun Shen prepared and scanned renal pathology samples. Yong Wang and Xueying Cao were involved in patient enrollment. Guangyan Cai provided platform support and contributed to manuscript revision. Xiangmei Chen, Haiyi Wang, and Zheyi Dong contributed to study design and manuscript revision. All authors read and approved the final manuscript. Each author has participated sufficiently in this work to take public responsibility for the content. All authors have contributed to the conception, design, execution, analysis, and interpretation of the study. All authors have reviewed and approved the final version of the manuscript. Funding This research was supported by the Capital’s Funds for Health Improvement and Research (No. CFH 2024-1-5021), the Science & Technology Project of Beijing, China (No. Z221100007422121), the Beijing Natural Science Foundation (Nos. L222133, L232122, and 7252018), the Noncommunicable Chronic Diseases-National Science and Technology Major Project (Nos. 2026ZD0557100, 2026ZD0557105) and the National Natural Science Foundation (Nos. U24A20755 and 62450131). Data availability The datasets generated and analyzed during the present study are available from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate This study was approved by the Research Ethics Committee of the Chinese People’s Liberation Army General Hospital (no. S2017-133-01). Consent for publication Informed consent was obtained from all participants. Competing interests L.X. is affiliated with GE Healthcare. The remaining authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Qian Wang, Shaopeng Zhou, and Yue Niu contributed equally to this work. Contributor Information Xiangmei Chen, Email: [email protected]. Haiyi Wang, Email: [email protected]. Zheyi Dong, Email: [email protected]. Supplementary information The online version contains supplementary material available at 10.1186/s13244-026-02274-9. References 1. Molitch ME, Adler AI, Flyvbjerg A et al (2015) Diabetic kidney disease: a clinical update from kidney disease: improving global outcomes. Kidney Int 87:20–30. 10.1038/ki.2014.128 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Selby NM, Taal MW (2020) An updated overview of diabetic nephropathy: diagnosis, prognosis, treatment goals and latest guidelines. Diabetes Obes Metab 22:3–15. 10.1111/dom.14007 [ DOI ] [ PubMed ] [ Google Scholar ] 3. DeFronzo RA, Reeves WB, Awad AS (2021) Pathophysiology of diabetic kidney disease: impact of SGLT2 inhibitors. Nat Rev Nephrol 17:319–334. 10.1038/s41581-021-00393-8 [ DOI ] [ PubMed ] [ Google Scholar ] 4. Umanath K, Lewis JB (2018) Update on diabetic nephropathy: core curriculum 2018. Am J Kidney Dis 71:884–895. 10.1053/j.ajkd.2017.10.026 [ DOI ] [ PubMed ] [ Google Scholar ] 5. American Diabetes Association (2018) 10. Microvascular complications and foot care: standards of medical care in diabetes-2018. Diabetes Care 41:S105–S118. 10.2337/dc18-S010 [ DOI ] [ PubMed ] [ Google Scholar ] 6. Zhang L, Long J, Jiang W et al (2016) Trends in chronic kidney disease in China. N Engl J Med 375:905–906. 10.1056/NEJMc1602469 [ DOI ] [ PubMed ] [ Google Scholar ] 7. Johansen KL, Garimella PS, Hicks CW et al (2021) Central and peripheral arterial diseases in chronic kidney disease: conclusions from a kidney disease: improving Global Outcomes (KDIGO) Controversies Conference. Kidney Int 100:35–48. 10.1016/j.kint.2021.04.029 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 8. Jiang W, Wang J, Shen X et al (2020) Establishment and validation of a risk prediction model for early diabetic kidney disease based on a systematic review and meta-analysis of 20 cohorts. Diabetes Care 43:925–933. 10.2337/dc19-1897 [ DOI ] [ PubMed ] [ Google Scholar ] 9. Heyman SN, Raz I, Dwyer JP, Weinberg Sibony R, Lewis JB, Abassi Z (2022) Diabetic proteinuria revisited: updated physiologic perspectives. Cells 11:2917. 10.3390/cells11182917 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 10. Oshima M, Shimizu M, Yamanouchi M et al (2021) Trajectories of kidney function in diabetes: a clinicopathological update. Nat Rev Nephrol 17:740–750. 10.1038/s41581-021-00462-y [ DOI ] [ PubMed ] [ Google Scholar ] 11. Ebrahimi B, Textor SC, Lerman LO (2014) Renal relevant radiology: renal functional magnetic resonance imaging. Clin J Am Soc Nephrol 9:395–405. 10.2215/CJN.02900313 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Wang R, Lin Z, Yang X et al (2022) Noninvasive evaluation of renal hypoxia by multiparametric functional MRI in early diabetic kidney disease. J Magn Reson Imaging 55:518–527. 10.1002/jmri.27814 [ DOI ] [ PubMed ] [ Google Scholar ] 13. Liang P, Yuan G, Li S et al (2023) Noninvasive assessment of the renal function, Oxford classification and prognostic risk stratification of IgAN by using intravoxel incoherent motion diffusion-weighted imaging and blood oxygenation level-dependent MRI. J Magn Reson Imaging 58:879–891. 10.1002/jmri.28565 [ DOI ] [ PubMed ] [ Google Scholar ] 14. Hesp AC, Schaub JA, Prasad PV et al (2020) The role of renal hypoxia in the pathogenesis of diabetic kidney disease: a promising target for newer renoprotective agents including SGLT2 inhibitors? Kidney Int 98:579–589. 10.1016/j.kint.2020.02.041 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Bane O, Seeliger E, Cox E et al (2023) Renal MRI: from nephron to NMR signal. J Magn Reson Imaging 58:1660–1679. 10.1002/jmri.28828 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Sugiyama K, Inoue T, Kozawa E et al (2020) Reduced oxygenation but not fibrosis defined by functional magnetic resonance imaging predicts the long-term progression of chronic kidney disease. Nephrol Dial Transplant 35:964–970. 10.1093/ndt/gfy324 [ DOI ] [ PubMed ] [ Google Scholar ] 17. Li LP, Thacker JM, Li W et al (2020) Medullary blood oxygen level-dependent MRI index (R2*) is associated with annual loss of kidney function in moderate CKD. Am J Nephrol 51:966–974. 10.1159/000512854 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Feng YZ, Ye YJ, Cheng ZY et al (2020) Non-invasive assessment of early stage diabetic nephropathy by DTI and BOLD MRI. Br J Radiol 93:20190562. 10.1259/bjr.20190562 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. Levey AS, Stevens LA, Schmid CH et al (2009) A new equation to estimate glomerular filtration rate. Ann Intern Med 150:604–612. 10.7326/0003-4819-150-9-200905050-00006 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Wang W, Yu Y, Wen J et al (2019) Combination of functional magnetic resonance imaging and histopathologic analysis to evaluate interstitial fibrosis in kidney allografts. Clin J Am Soc Nephrol 14:1372–1380. 10.2215/CJN.00020119 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Tervaert TWC, Mooyaart AL, Amann K et al (2010) Pathologic classification of diabetic nephropathy. J Am Soc Nephrol 21:556–563. 10.1681/ASN.2010010010 [ DOI ] [ PubMed ] [ Google Scholar ] 22. Piskunowicz M, Hofmann L, Zuercher E et al (2015) A new technique with high reproducibility to estimate renal oxygenation using BOLD-MRI in chronic kidney disease. Magn Reson Imaging 33:253–261. 10.1016/j.mri.2014.12.002 [ DOI ] [ PubMed ] [ Google Scholar ] 23. Alicic RZ, Rooney MT, Tuttle KR (2017) Diabetic kidney disease: challenges, progress, and possibilities. Clin J Am Soc Nephrol 12:2032–2045. 10.2215/CJN.11491116 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 24. Francis ST, Selby NM, Taal MW (2023) Magnetic resonance imaging to evaluate kidney structure, function, and pathology: moving toward clinical application. Am J Kidney Dis 82:491–504. 10.1053/j.ajkd.2023.02.007 [ DOI ] [ PubMed ] [ Google Scholar ] 25. Xu Y, Yang J, Lu F, Ye C, Wang C (2023) Correlation of renal oxygenation with renal function in chronic kidney disease: a preliminary prospective study. Kidney Blood Press Res 48:175–185. 10.1159/000529165 [ DOI ] [ PubMed ] [ Google Scholar ] 26. Inoue T, Kozawa E, Ishikawa M et al (2021) Comparison of multiparametric magnetic resonance imaging sequences with laboratory parameters for prognosticating renal function in chronic kidney disease. Sci Rep 11:22129. 10.1038/s41598-021-01147-z [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Sørensen SS, Gullaksen S, Vernstrøm L et al (2023) Evaluation of renal oxygenation by BOLD-MRI in high-risk patients with type 2 diabetes and matched controls. Nephrol Dial Transplant 38:691–699. 10.1093/ndt/gfac186 [ DOI ] [ PubMed ] [ Google Scholar ] 28. Chen F, Yan H, Yang F et al (2021) Evaluation of renal tissue oxygenation using blood oxygen level-dependent magnetic resonance imaging in chronic kidney disease. Kidney Blood Press Res 46:441–451. 10.1159/000515709 [ DOI ] [ PubMed ] [ Google Scholar ] 29. Li C, Liu H, Li X, Zhou L, Wang R, Zhang Y (2019) Application of BOLD-MRI in the classification of renal function in chronic kidney disease. Abdom Radiol (NY) 44:604–611. 10.1007/s00261-018-1750-6 [ DOI ] [ PubMed ] [ Google Scholar ] 30. Yin WJ, Liu F, Li XM et al (2012) Noninvasive evaluation of renal oxygenation in diabetic nephropathy by BOLD-MRI. Eur J Radiol 81:1426–1431. 10.1016/j.ejrad.2011.03.045 [ DOI ] [ PubMed ] [ Google Scholar ] 31. Ricciardi CA, Gnudi L (2021) Kidney disease in diabetes: from mechanisms to clinical presentation and treatment strategies. Metabolism 124:154890. 10.1016/j.metabol.2021.154890 [ DOI ] [ PubMed ] [ Google Scholar ] 32. Fan X, Yang M, Lang Y et al (2024) Mitochondrial metabolic reprogramming in diabetic kidney disease. Cell Death Dis 15:442. 10.1038/s41419-024-06833-0 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 33. Wang ZJ, Kumar R, Banerjee S, Hsu CY (2011) Blood oxygen level-dependent (BOLD) MRI of diabetic nephropathy: preliminary experience. J Magn Reson Imaging 33:655–660. 10.1002/jmri.22501 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 34. Seah JM, Botterill E, MacIsaac RJ, Milne M, Ekinci EI, Lim RP (2022) Functional MRI in assessment of diabetic kidney disease in people with type 1 diabetes. J Diabetes Complications 36:108076. 10.1016/j.jdiacomp.2021.108076 [ DOI ] [ PubMed ] [ Google Scholar ] 35. He X, Deng Y, Tian B, Zhao Y, Han M, Cai Y (2024) A retrospective cohort study of clinical characteristics and outcomes of type 2 diabetic patients with kidney disease. PeerJ 12:e16915. 10.7717/peerj.16915 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 36. Packer M (2021) Mechanisms leading to differential hypoxia-inducible factor signaling in the diabetic kidney: modulation by SGLT2 inhibitors and hypoxia mimetics. Am J Kidney Dis 77:280–286. 10.1053/j.ajkd.2020.04.016 [ DOI ] [ PubMed ] [ Google Scholar ] 37. Yamanouchi M, Furuichi K, Shimizu M et al (2022) Serum hemoglobin concentration and risk of renal function decline in early stages of diabetic kidney disease: a nationwide, biopsy-based cohort study. Nephrol Dial Transplant 37:489–497. 10.1093/ndt/gfab185 [ DOI ] [ PubMed ] [ Google Scholar ] 38. Collins GS, Reitsma JB, Altman DG, Moons KGM (2015) Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ 350:g7594. 10.1136/bmj.g7594 [ DOI ] [ PubMed ] [ Google Scholar ] 39. Ho J, Tumkaya T, Aryal S, Choi H, Claridge-Chang A (2019) Moving beyond P values: data analysis with estimation graphics. Nat Methods 16:565–566. 10.1038/s41592-019-0470-3 [ DOI ] [ PubMed ] [ Google Scholar ] 40. Ioannidis JPA (2019) The importance of predefined rules and prespecified statistical analyses: do not abandon significance. JAMA 321:2067–2068. 10.1001/jama.2019.4582 [ DOI ] [ PubMed ] [ Google Scholar ] 41. Breiman L (2001) Random forests. Mach Learn 45:5–32. 10.1023/A:1010933404324 [ Google Scholar ] 42. Pruijm M (2020) Can COMBINED magnetic resonance imaging measure the progression of kidney disease? Clin J Am Soc Nephrol 15:747–749. 10.2215/CJN.04430420 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials ELECTRONIC SUPPLEMENTARY MATERIAL (248.9KB, pdf) Data Availability Statement The datasets generated and analyzed during the present study are available from the corresponding author upon reasonable request. Articles from Insights into Imaging are provided here courtesy of Springer ACTIONS View on publisher site PDF (3.6 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top