FUTURE DIRECTIONS AND RESEARCH GAPS

The future of kidney imaging lies in its integration with genetics, clinical biomarkers, AI, and structural and functional data to deliver precision medicine. Emerging techniques such as multiparametric MRI [66, 67], US-based radiomics [13], and H-scan ultrasonography represent promising alternatives to biopsy and conventional laboratory tests. However, several unresolved challenges must be addressed before these methods can be widely implemented in clinical nephrology (Table 2).

Disease | Imaging and quantitative read-outs | Utility and incremental value | Limitations | Research gaps and recommendations
CKD | ASL → RBF; BOLD → R2*; DWI/IVIM → ADC, f, D; native T1 mapping; Doppler → RRI; SWE/MRE → stiffness | BOLD R2* (↓ oxygenation) predicts faster eGFR decline, adverse outcomes [30].ASL-RBF falls stepwise with CKD stage, discriminates CKD vs. controls [34].IVIM (f, D) and native T1 associate with fibrosis/progression beyond routine markers [40, 77]. | BOLD influenced by hemodynamics/blood volume unless corrected; between-study heterogeneity [33].US measures (Doppler/SWE) are operator-dependent; inter-vendor variability for MR sequences [19]. | Multicenter harmonization of ASL/IVIM/T1/BOLD protocols & phantoms; embed ΔR2*, ΔRBF, ΔT1 as prespecified endpoints; integrate imaging with omics/ML for calibrated risk prediction [78].
DKD | ASL/PC-MRI → RBF; BOLD → R2*; DWI → cortical ADC; native T1 mapping; Doppler → RRI; SWE → cortical stiffness; multiparametric non-contrast MRI | Cortical ADC stratifies DKD stage and predicts outcomes beyond eGFR/UACR [79].SGLT2 inhibitors acutely increase cortical oxygenation on BOLD-MRI (early pharmacodynamic read-out) [80].Multiparametric MRI feasible for DKD phenotyping without gadolinium [81]. | Many DKD studies are single-center; cross-scanner cut-offs not harmonized; US indices vary with technique [81]. | Validate reproducible ADC/T1 thresholds across vendors; use multiparametric MRI as response markers in DKD trials; fuse imaging with urinary biomarkers (TNFR-1/KIM-1) and polygenic scores in ML models [78].
HKD | ASL → RBF; IVIM → f, D; T1 mapping; BOLD → R2*; Doppler → RRI; SWE → cortical stiffness; CEUS → perfusion index/flow reserve; time-SLIP MRA → RAS grading (non-contrast) | CEUS shows reduced cortical perfusion and decreased renal flow reserve in hypertension [18].RRI ≥0.70 associates with higher mortality and CKD progression risk [82].Time-SLIP non-contrast MRA accurately detects/grades renal artery stenosis (multicenter) [47]. | CEUS/US are operator-dependent; BOLD requires hemodynamic context; limited biopsy-linked validation specific to hypertension [19, 33] | Standardise CEUS metrics and test imaging-guided therapy titration (e.g. RAAS/diuretics); expand non-contrast MRA validation in CKD cohorts [83, 84].
Fatty kidney/metabolic renal phenotype | Dixon MRI → PDFF; MRS → intrarenal lipid content; BOLD → R2*; CT → RSF (volumetric/APDRS); automated CT segmentation (DL) | Highest RSF decile linked to higher odds of hypertension and CKD independent of visceral fat [50].APDRS correlates with volumetric RSF and shows excellent reproducibility [48].Semaglutide/empagliflozin: early MRI signals (↓cortical ADC, ↓TKV) over 32 weeks [55]. | No harmonised PDFF/RSF pathological cut-offs; vendor/sequence effects; CT radiation; causality unresolved [43]. | Define PDFF/RSF thresholds across scanners/ethnicities; validate automated segmentation/radiomics across centres; test whether ΔPDFF/ΔRSF independently predict hard renal outcomes [48, 55].
Kidney allograft (transplant) | Multiparametric MRI → DWI/ASL/BOLD/T1/T2; Doppler → RRI trajectories; CEUS → cortical perfusion; SWE → cortical stiffness; MRE → whole-kidney stiffness; H-scan → fibrosis index; PAI (during NMP) → cortical sO₂ | RI early/serial predicts delayed graft function and adverse outcomes [85].SWE shows moderate correlation with Banff fibrosis (systematic review/meta-analysis) [86].H-scan (first-in-human) correlates with 12-mo eGFR; PAI during NMP discriminates functional vs. non-functional grafts by cortical oxygenation [87, 88]. | Inconsistent imaging–histology concordance (esp. MRE); heterogeneity in SWE thresholds; clinical-decision cut-offs lacking; limited multi-center standardization [86, 89]. | Prospective imaging–biopsy correlation with standardized protocols; evaluate ΔT1/Δperfusion/ΔSWE as surrogate endpoints; integrate AI/radiomics into transplant registries; expand CEUS/PAI/H-scan validation across centres [90].

Limitations and cause of variation

Several obstacles today impede the broad clinical use of quantitative renal imaging biomarkers. Technical restrictions include differences between scanners and suppliers, differences in acquisition methodologies and post-processing pipelines, and the lack of universally recognized thresholds for modalities including as SWE, BOLD MRI, ASL, and native T1 mapping. Biological variables such as hydration state, blood pressure, anemia, and concurrent cardiovascular sickness all have an impact on imaging results and complicate cross-study comparisons. Lastly, implementation barriers include cost, scanner availability, a lack of nephrology-focused imaging expertise, and the lack of support from guidelines hinder routine clinical use. In addition, while many studies report promising associations with fibrosis, progression, or outcomes, negative or equivocal findings and modest effect sizes have also been described, underscoring the need for larger, multicenter, and longitudinal validation before imaging biomarkers can be widely used as surrogate endpoints [3, 67, 68].

Validation and generalizability. A key limitation is the lack of longitudinal, multicenter validation studies, including repeated T1 mapping or ASL measurements within established CKD cohorts. Most imaging biomarkers have been validated only in small, often cross-sectional datasets, which undermines their prognostic value and generalizability [67, 68]. For instance, although T1 mapping and DWI correlate with fibrosis and declining kidney function [68, 69], their ability to reliably predict therapeutic response in DKD or across CKD stages remains uncertain.

Clinical trial design

Another gap lies in the limited use of imaging as primary or secondary endpoints in clinical trials. To enable imaging-guided staging, stratification, and therapeutic monitoring, trial designs must be reconfigured. While regulatory bodies such as the United States Food and Drug Administration and European Medicines Agency have expressed openness toward imaging-based surrogate endpoints, standardized acquisition protocols and analytic reproducibility are prerequisites [3, 70].

Multi-omics and AI integration

Future research should also advance the integration of imaging with transcriptomic, proteomic, metabolomic, and genetic datasets. This multimodal approach could generate high-resolution patient phenotypes, redefine clinical staging, and individualize therapy. Urinary proteomic classifiers (e.g. CKD273), metabolomic signatures, and polygenic risk scores have shown promise in predicting accelerated eGFR loss and cardiovascular risk beyond standard variables. Combining these molecular readouts with quantitative imaging markers—such as cortical T1, ASL perfusion, or elastography-based stiffness—may substantially refine risk stratification in kidney disease, although prospective validation is still required [71, 72]. The integration of such high-dimensional data through AI and machine learning offers further opportunities, though barriers remain, including real-time incorporation into electronic health records, external validation, and model interpretability [71, 73].

A notable shortcoming exists in the utilization of imaging for monitoring treatment interventions. While SGLT2 inhibitors, MRAs, and anti-fibrotic agents show promise in slowing CKD progression, there is less data regarding the use of imaging as an objective, non-invasive measure of therapy response. Future research may enhance treatment decision-making by evaluating how changes in imaging measures (e.g. cortical T1, renal perfusion) relate to therapy outcomes [69, 74].

Ultimately, implementation obstacles—such as financial constraints, accessibility issues, insufficient nephrology-focused training in imaging interpretation, and the lack of inclusion in guidelines—further impede practical application. National initiatives, such as the UK’s NURTuRE imaging substudy [75], and European consortia like PARENCHIMA (3) are commencing efforts to overcome these obstacles via standardization and multicenter validation.

In conclusion, although kidney imaging is set to revolutionize prognostication and treatment oversight in nephrology, achieving this potential necessitates focused research in five critical domains: (i) longitudinal and multicenter validation [67, 70], (ii) imaging-informed clinical trials [69], (iii) multi-omics integration and AI-driven modeling [13, 73], (iv) therapeutic monitoring endpoints [69, 74], and (v) implementation science to enhance guideline incorporation and equitable access [3, 75].

Integrating these high-dimensional imaging maps with emerging fluid biomarkers such as urinary tumor necrosis factor receptor 1 (TNFR-1) and kidney injury molecule 1 (KIM-1), as well as polygenic risk scores—and interrogating the ensemble through machine learning—hints at a future where DKD is staged not by creatinine and albuminuria but by truly multidimensional, actionable endotypes. Prospective multicenter trials are now needed to validate these imaging biomarkers as surrogate endpoints and to assess their cost-effectiveness in precision nephrology.

Clinical translation and practical implications

From a clinical perspective, several renal imaging tools discussed in this review are already suitable for implementation in routine practice, while others remain investigational. Currently implementable approaches include Doppler-derived renal resistive index and ultrasound elastography, which can be used as supportive markers of intrarenal vascular resistance and fibrosis in selected contexts such as CKD progression assessment, hypertensive nephrosclerosis, and kidney transplant surveillance, particularly when interpreted longitudinally rather than as isolated measurements [19, 21, 25, 29]. Promising but still research-grade modalities include multiparametric MRI techniques—such as native T1 mapping, ASL-derived perfusion, and diffusion-based metrics—and their radiomics extensions, which have demonstrated strong associations with fibrosis, hypoxia, and outcomes but currently lack standardized thresholds, multicenter validation, and guideline endorsement [3, 32, 67]. Finally, primarily experimental techniques, including photoacoustic oxygen mapping, H-scan ultrasound-derived fibrosis indices, and advanced AI-driven multimodal models, remain confined to specialized research settings and early-phase clinical studies [12, 64, 73]. Explicitly distinguishing these tiers may help clinicians integrate available imaging tools into current care pathways while anticipating future developments as evidence matures.

As summarized in Fig. 4, a pragmatic, tiered imaging pathway is proposed in which widely available ultrasound-based techniques serve as first-line tools for routine CKD assessment, while advanced modalities such as CEUS, multiparametric MRI, and CT/MRA are reserved for selected clinical indications. Emerging technologies, including AI-assisted imaging, radiomics, PAI, and imaging–multi-omics, are currently best suited for research and tertiary care settings, reflecting the need to balance diagnostic value with resource availability and clinical complexity.

Figure 4:: A stepwise imaging pathway for CKD based on clinical complexity and resource availability. Figure legend: Proposed framework for integrating imaging into CKD assessment according to clinical indication, expertise, and resource requirements. Routine evaluation begins with renal ultrasonography, Doppler ultrasound/RRI, and elastography where available. Patients with uncertain findings or higher-risk disease may be referred for CEUS, mpMRI, or CT/MRA. Emerging techniques, including AI, radiomics, PAI, and imaging–multi-omics, are primarily intended for research and tertiary centers. Imaging should be selected according to the clinical question, local expertise, and availability. Abbreviations: AI, artificial intelligence; ASL, arterial spin labeling; BOLD, blood oxygen level-dependent; CEUS, contrast-enhanced ultrasound; CKD, chronic kidney disease; CT, computed tomography; DWI, diffusion-weighted imaging; ADC, apparent diffusion coefficient; mpMRI, multiparametric magnetic resonance imaging; MRA, magnetic resonance angiography; PDFF, proton density fat fraction; PRS, polygenic risk score; RRI, renal resistive index; UACR, urine albumin-to-creatinine ratio; and US, ultrasound.