, Introduction
Kidney transplantation is globally recognized as the treatment of choice for end-stage renal disease. Compared with maintenance dialysis, successful renal transplantation not only significantly reduces all-cause mortality but also substantially improves patients’ quality of life and facilitates the restoration of social functioning [1]. Driven by the maturation of surgical techniques, the optimization of organ preservation solutions, and the widespread application of novel immunosuppressive agents, short-term outcomes for recipients have reached historical highs where 1-year graft survival rates routinely exceed 90% [2]. Nevertheless, ensuring long-term allograft survival remains a formidable challenge as 10-year graft loss rates remain persistently elevated [3]. The pivotal factor underlying this attrition is immune-mediated injury, particularly chronic active antibody-mediated rejection (ABMR) and subclinical T-cell–mediated rejection (TCMR), which insidiously erode graft function and ultimately culminate in irreversible fibrosis and graft failure [4].
Historically, the diagnosis of renal allograft rejection has relied primarily on the monitoring of serum creatinine (sCr) and invasive allograft biopsy [5]. However, this traditional diagnostic framework appears increasingly inadequate in the era of precision medicine as it is plagued by inherent limitations. First, the delayed nature and low sensitivity of functional biomarkers represent the primary challenge in clinical monitoring. As a surrogate marker for estimated glomerular filtration rate (eGFR), sCr levels are influenced by various non-renal factors including muscle mass, dietary intake, drug interference, and tubular secretion [6]. More importantly, the kidney possesses a robust compensatory reserve capacity where significant elevations in sCr often occur only after substantial structural damage has affected more than 50% of the nephrons [7]. This implies that by the time diagnostic procedures are initiated based on elevated creatinine, the allograft has often already suffered irreversible structural damage and the optimal window for immune intervention has been missed. Second, the risks and limitations associated with invasive biopsy restrict its feasibility as a routine monitoring tool. Although the Banff classification system provides a standardized framework for the pathological diagnosis of rejection [8], renal biopsy remains an inherently invasive procedure that carries risks of hemorrhage, arteriovenous fistula formation, infection, and even graft loss due to complications [9]. These potential risks make clinicians reluctant to perform frequent surveillance biopsies on stable patients and usually restrict the procedure to for-cause biopsies triggered by clinical indications such as elevated creatinine or increased proteinuria. Furthermore, biopsy cores typically contain fewer than 20 glomeruli representing a minute fraction of approximately 1/10,000 of the total renal tissue which leads to significant sampling error [10]. In cases of patchily distributed lesions such as early focal inflammation or BK virus nephropathy, the biopsy needle may miss the affected area and result in a false negative diagnosis [11]. Moreover, pathological diagnosis relies heavily on the subjective experience of the pathologist, and inter-observer variability remains a persistent challenge confounding the Banff classification system [12].
In light of these challenges, there is an urgent need within the transplant medicine community for a novel diagnostic paradigm that enables early warning, allows for high-frequency monitoring, and ensures safety through a noninvasive approach. Driven by rapid advancements in high-throughput sequencing, molecular biology, and functional imaging in recent years, noninvasive diagnosis has transitioned from laboratory research to clinical translation and has become a focal point for international authoritative guidelines such as those issued by KDIGO (Kidney Disease: Improving Global Outcomes) and the European Society for Organ Transplantation (ESOT) [13,14]. Current frontier advancements in the noninvasive diagnosis of renal allograft rejection are primarily advancing through the synergistic application of two complementary technological pathways. First, liquid biopsy technologies enable the dynamic and real-time monitoring of immune activity within the allograft by utilizing high-throughput analysis to detect biomarkers such as cell-free DNA (cfDNA), RNA, proteins, and exosomes present in body fluids including blood and urine. This approach is capable of sensitively capturing rejection-related molecular signals before the onset of overt structural damage to the tissue [15]. Second, significant breakthroughs in functional imaging technologies such as ultrasound elastography, contrast-enhanced ultrasound (CEUS), and multiparametric magnetic resonance imaging (mpMRI) now allow for the quantitative assessment of renal physical properties at a macroscopic level. These properties include tissue stiffness, microvascular perfusion, water diffusion capacity, and local oxygenation status which collectively enable the spatial localization and visual assessment of the pathophysiological state of the graft [16]. Together, these two directions constitute a multi-level and noninvasive diagnostic system bridging molecular mechanisms with macroscopic phenotypes.
This review systematically summarizes the latest research findings in these fields with a focus on technical principles, clinical validation data, and future application prospects to provide a reference for ongoing research and clinical practice regarding the noninvasive diagnosis of renal allograft rejection. This narrative review is based on a structured literature search conducted in PubMed and Web of Science. Studies published primarily within the past decade were prioritized, with additional inclusion of earlier landmark studies where appropriate. Emphasis was placed on high-quality evidence, including prospective and multicenter studies, as well as clinically validated diagnostic approaches. Emerging technologies with potential translational relevance were also selectively included. To visually illustrate this novel paradigm integrating molecular messengers, functional imaging, and artificial intelligence (AI), a schematic overview is provided in Fig. 1.
Liquid biopsy: a microscopic window into genomics and immunology
The core philosophy underlying liquid biopsy involves the utilization of circulating biomarkers within body fluids to infer the health status of an organ. In the field of renal transplantation, donor-derived cfDNA (dd-cfDNA) and urinary chemokines have emerged as the biomarkers with the greatest prospects for clinical application as they respectively represent the distinct pathophysiological dimensions of cell injury and immune recruitment. Table 1 summarizes and compares the biological characteristics and clinical utility of three major validated liquid biopsy technologies.
Donor-derived cell-free DNA
Biological mechanisms and release kinetics
The cfDNA consists of double-stranded DNA fragments released into the systemic circulation during the processes of cell apoptosis or necrosis. These fragments typically display a length distribution ranging from 150 to 200 base pairs (bp) which reflects the characteristic feature of nucleosome protection [17]. In renal transplant recipients, circulating cfDNA comprises two distinct components. The vast majority originates from the turnover of the recipient’s own cells, particularly those of the hematopoietic system, and is termed recipient-derived cfDNA. Conversely, a minute fraction is derived from the death of grafted kidney cells and is referred to as dd-cfDNA [18]. Under stable conditions characterized by immune quiescence, the turnover rate of donor renal cells is minimal and the fraction of dd-cfDNA within the total cfDNA pool typically remains below 1% or even falls under 0.5%. However, during episodes of rejection encompassing both ABMR and TCMR, immune-mediated attacks induce extensive apoptosis and necrosis in allograft cells such as endothelial cells and renal tubular epithelial cells. This injury precipitates a sharp increase in the rate of dd-cfDNA release into the bloodstream [19]. A distinct advantage of dd-cfDNA lies in its rapid kinetic profile. The half-life of cfDNA in the blood is extremely short and ranges from only 30 to 60 minutes as it is cleared primarily through hepatic metabolism and renal excretion [20]. Consequently, dd-cfDNA is capable of reflecting the current injury status of the allograft in near real-time. In contrast, elevations in sCr often require days or even weeks to accumulate and resolve slowly even after the cessation of injury. Functioning as a real-time injury marker, changes in dd-cfDNA levels can be utilized to monitor the efficacy of antirejection therapies where effective treatment is typically signaled by a rapid decline in dd-cfDNA levels [21].
Detection technologies
Currently, the primary technical challenge in detecting trace amounts of dd-cfDNA in the blood lies in achieving high-sensitivity differentiation against the substantial background noise of recipient DNA. Mainstream technologies employed for this purpose primarily rely on two methodological approaches including targeted next-generation sequencing and droplet digital polymerase chain reaction (PCR). Targeted next-generation sequencing represents the most widely applied commercial pathway and operates based on genotypic disparities between the donor and recipient at single-nucleotide polymorphism (SNP) loci. This technique involves designing a pool of probes covering hundreds to thousands of high-frequency heterozygous SNP sites to perform targeted enrichment and sequencing of cfDNA in plasma [22]. Subsequently, bioinformatics algorithms analyze the allele frequencies within the sequencing reads to accurately calculate the proportion of donor alleles. A significant advantage of this method lies in its ability to function without the prior acquisition of donor genetic samples which makes it particularly suitable for scenarios such as deceased donor transplantation. Additionally, it features high-throughput capabilities that allow for the parallel processing of large sample volumes. However, the turnaround time for detection typically spans 2 to 3 days, costs are relatively high, and the process imposes substantial demands on downstream bioinformatics analysis capabilities [23].
Another pivotal technology is droplet digital PCR. This method utilizes microfluidic technology to partition the PCR reaction system into tens of thousands of nanoliter-sized droplets so that each droplet contains either zero or one target DNA molecule. Through endpoint fluorescence detection of all droplets, the absolute copy number of the target molecules is calculated precisely based on the principles of the Poisson distribution [24]. This technology boasts extremely high sensitivity capable of detecting dd-cfDNA fractions as low as 0.1% while providing results within hours at a relatively lower cost. Furthermore, it can simultaneously determine both the fractional abundance and the absolute concentration of dd-cfDNA [25]. Its limitation lies in the general requirement to design primers for specific high-frequency insertion-deletions or SNP sites which may depend on known donor and recipient genotype information or require the use of universal multi-locus kits.
Clinical validation and diagnostic efficacy
The clinical validity of dd-cfDNA has been robustly supported by multiple large-scale multicenter prospective studies which have established it as a pivotal noninvasive biomarker for the diagnosis of allograft rejection. The DART study stands as a landmark investigation establishing the diagnostic value of dd-cfDNA in differentiating types of acute rejection (AR). Data from this study demonstrated the exceptional performance of dd-cfDNA in identifying ABMR where median levels in patients with ABMR reached 2.9% significantly exceeding the 1.2% observed in patients with TCMR and the 0.3% found in the non-rejection control group [26]. A subsequent meta-analysis pooling data from 12 studies further confirmed this finding by reporting a pooled area under the curve (AUC) of 0.87 for the diagnosis of ABMR with sensitivity and specificity reaching 0.81 and 0.80, respectively [27]. In contrast, the diagnostic efficacy of dd-cfDNA for TCMR is relatively limited, particularly for low-grade lesions where the AUC ranges from approximately 0.71 to 0.73. This disparity may be attributed to the fact that TCMR primarily causes focal tubular injury resulting in a lower quantity of DNA released into the blood whereas ABMR directly damages the vascular endothelium leading to a greater influx of DNA into the circulation.
Beyond immediate diagnosis, dd-cfDNA has also demonstrated significant value in predicting long-term allograft function. The ADMIRAL study longitudinally monitored 1,092 patients and found that dd-cfDNA levels exceeding 0.5% were significantly associated with subsequent declines in eGFR. Persistently elevated dd-cfDNA levels nearly doubled the risk of a 25% decline in eGFR within 3 years posttransplantation which supports its utility as a long-term monitoring tool for identifying patients at high risk of subclinical injury [28]. The TRIFECTA study provided critical evidence regarding the biological basis of dd-cfDNA by prospectively collecting a large scale of paired samples. This study confirmed for the first time that plasma dd-cfDNA levels correlate strongly with molecular rejection phenotypes found in allograft biopsies, particularly in cases of active ABMR and mixed rejection where the association was even superior to traditional histological assessment [29]. This marks the evolution of dd-cfDNA from a mere marker of injury to a biological indicator directly linked to immune rejection activity. Recent large-scale multicenter evidence has further strengthened the clinical utility of dd-cfDNA. A large observational study demonstrated that dd-cfDNA is independently associated with allograft rejection and significantly improves the discriminative performance of standard clinical models, with AUC increasing from 0.777 to 0.821 [30]. Beyond its role as a standalone biomarker, subsequent studies have explored its application in clinical stratification and highlighted the value of integrating dd-cfDNA with immune markers. For example, a prospective study showed that combining dd-cfDNA at a threshold greater than 0.4% with donor-specific antibodies significantly increased the specificity for diagnosing ABMR to 90.7% [31]. Collectively, these findings support the incorporation of dd-cfDNA into multiparametric diagnostic frameworks to enhance the precision of clinical assessment.
Critical controversies and frontier advances
Although the reporting of dd-cfDNA results currently relies primarily on the fraction of total cfDNA as a core metric with clinical cutoffs typically set at 1.0% or 0.5%, this fractional index suffers from inherent mathematical limitations [32]. As a relative ratio, its value is influenced not only by the numerator representing the quantity of dd-cfDNA molecules derived from allograft injury but is also highly susceptible to interference from fluctuations in the denominator representing the recipient’s own cfDNA levels. For instance, during episodes of infection or physiological stress in the recipient, a surge in cfDNA released by autologous leukocytes can dilute the dd-cfDNA fraction and lead to false-negative results, whereas a reduction in total cfDNA in the context of leukopenia can cause a spurious elevation in the fraction and result in false positives [33]. To address this deficiency, the academic community is increasingly advocating for the adoption of absolute quantification as a complementary or alternative metric. Studies indicate that measuring the absolute concentration of dd-cfDNA offers superior sensitivity compared to using the fraction alone in certain scenarios [34]. Furthermore, a recent study proposed an innovative dual-positive diagnostic strategy which requires both a dd-cfDNA fraction exceeding 1.08% and an absolute copy number greater than 32 cp/mL. This strategy elevated diagnostic specificity to 97.14% and significantly optimized the positive predictive value which suggests that future clinical interpretation standards may evolve toward this composite model [35].
It is imperative to clarify that dd-cfDNA serves fundamentally as a marker of allograft injury rather than a specific indicator of rejection. Any pathological process resulting in allograft cell injury or death including BK virus nephropathy, acute tubular necrosis (ATN), or severe systemic infection may precipitate an elevation in its levels [36]. Consequently, its primary clinical utility resides in its exceptionally high negative predictive value which allows it to serve as a reliable exclusionary tool. When dd-cfDNA levels remain low, clinicians can proceed with greater confidence in avoiding unnecessary biopsies. Conversely, elevated levels necessitate a comprehensive assessment integrated with donor-specific antibody status, clinical presentation, and other noninvasive biomarkers while renal biopsy remains requisite for the definitive determination of etiology when indicated. In addition, methodological factors should be considered when interpreting dd-cfDNA results. Differences between fractional and absolute quantification approaches may influence clinical interpretation, as these metrics reflect distinct biological aspects and may be differentially affected by systemic conditions. Furthermore, variability in cutoff values across platforms and studies remains an important limitation, which may hinder standardization and broader clinical implementation.
Another limitation of dd-cfDNA worth considering is that the interpretability of results is reduced in cases of multi-organ transplantation from the same donor. In such situations, currently available detection methods cannot distinguish the tissue origin of DNA fragments derived from the donor [37]. Consequently, elevated dd-cfDNA levels reflect the total burden of injury across all transplanted organs rather than providing organ-specific information. In the presence of multiple allografts, this limitation significantly restricts its clinical utility in diagnosing renal rejection. Emerging research approaches may help address this challenge. For example, recent studies have explored tissue-specific DNA methylation patterns, fragment size profiles, and nucleosome localization characteristics to achieve more precise attribution of cfDNA to its source organ [38]. Although these techniques are currently primarily in the research phase, they hold promise for improving the specificity and clinical utility of dd-cfDNA in complex transplant settings.
Urinary chemokines: an immune barometer for noninvasive monitoring
While blood provides a reflection of the aggregate systemic state, urine serves as a direct product of the renal microenvironment. Immune molecules present in urine originate directly from renal filtration and local secretion which confers an inherent advantage in reflecting local renal inflammation. In this context, the chemokines CXCL9 and CXCL10 have been validated as the most reliable urinary biomarkers for the diagnosis of rejection.
Immunobiology of the CXCL9/CXCL10-CXCR3 axis
CXCL9, also known as monokine induced by gamma interferon, and CXCL10, known as interferon-gamma-inducible protein 10 (IP-10), belong to the CXC chemokine family and exert their biological functions by binding to the CXCR3 receptor [39]. CXCR3 is predominantly expressed on the surface of activated Th1 cells, CD8+ cytotoxic T cells, and natural killer cells. During the initiation phase of the rejection cascade, antigen-presenting cells recognize alloantigens and subsequently activate T cells [40]. These activated T cells infiltrate the allograft tissue and release interferon-gamma (IFN-γ). Functioning as a potent pro-inflammatory cytokine, IFN-γ induces renal tubular epithelial cells and microvascular endothelial cells to synthesize and secrete copious amounts of CXCL9 and CXCL10 [41]. These chemokines are excreted into the urine while simultaneously establishing a concentration gradient that recruits additional CXCR3+ immune cells into the allograft, thereby creating a positive feedback loop of amplification that manifests as tubulitis and endothelitis [42]. Since CXCL9 and CXCL10 serve as critical mediators in the early phases of the immune response, their elevation in urine typically precedes alterations in sCr and the formation of histological damage, which confers significant early warning value to these biomarkers [43].
Diagnostic thresholds and clinical efficacy
Clinical guidelines issued by the ESOT recommend urinary chemokines as useful noninvasive tools in specific clinical contexts. In patients with stable graft function, urinary CXCL9 and CXCL10 can assist in excluding subclinical rejection, whereas in patients with graft dysfunction, they may serve as adjunctive biomarkers to support the diagnosis of AR [13].
The clinical utility of urinary chemokines has been supported by multiple lines of evidence. Early studies using urinary-cell messenger RNA (mRNA) profiling identified a three-gene signature consisting of CD3ε, CXCL10 (IP-10), and 18S ribosomal RNA for the noninvasive diagnosis of acute cellular rejection, highlighting the role of CXCL10 at the transcript level [44]. Subsequent clinical studies, including the CTOT-01 trial, further demonstrated that urinary CXCL9 protein levels can serve as an early diagnostic and predictive biomarker, with elevations preceding clinical rejection by several weeks [45].
In clinical practice, urinary chemokine levels are typically normalized to urinary creatinine to account for variations in urine concentration. Recent studies have reported that CXCL10/creatinine and CXCL9/creatinine ratios show moderate diagnostic performance, with AUC values around 0.7 and established cutoff values for clinical interpretation [46]. In addition to diagnostic utility, elevated urinary chemokine levels have been associated with adverse graft outcomes, including progression to graft dysfunction and long-term allograft loss.
Limitations and differential diagnosis
Similar to dd-cfDNA, the specificity of urinary chemokines is compromised by the presence of other inflammatory states. For instance, infections with the BK virus or cytomegalovirus activate robust Th1-type immune responses and induce the release of IFN-γ. This cascade leads to significantly elevated levels of urinary CXCL9 and CXCL10 which can sometimes exceed those observed in patients with allograft rejection [47]. Consequently, the interpretation of abnormally elevated urinary chemokines necessitates the prior exclusion of active viral infections via PCR testing. Furthermore, bacterial urinary tract infections including cystitis and pyelonephritis can also induce elevations in chemokine levels which requires a comprehensive differential diagnosis incorporating urinalysis and bacterial culture results to identify the underlying cause.
Extracellular vesicles and mRNA signatures
Extracellular vesicles, particularly exosomes, are nanoscale membranous vesicles with diameters ranging from 50 to 200 nm that are secreted by cells. Encapsulating cytosolic components from their parent cells such as mRNA, microRNA (miRNA), proteins, and metabolites, they are regarded as critical vehicles for intercellular communication [48].
Unique advantages of urinary exosomes
Urinary exosomes predominantly originate from glomerular structures including podocytes and endothelial cells as well as renal tubular epithelial cells and immune cells infiltrating the kidney [49]. Unlike cell-free RNA, intra-exosomal RNA is shielded by a lipid bilayer which confers exceptional stability by protecting it against degradation from high concentrations of urinary RNases [50]. Research has demonstrated that exosomal mRNA remains intact even when urine samples are stored at 4 °C for 2 weeks, a feature that significantly facilitates sample collection and transportation [51].
Exosomal mRNA gene signatures
Leveraging high-throughput sequencing and machine learning (ML) screening, researchers have successfully established multiple exosomal transcriptomic signatures with diagnostic utility. For instance, a urinary exosomal mRNA detection assay developed on the Thermo Fisher/Exosome Diagnostics platform demonstrated in a validation study involving 175 patients that a multigene mRNA signature derived from exosomal transcriptomic profiling correlates highly with biopsy-proven rejection [52]. This signature is capable not only of accurately distinguishing rejection from non-rejection states with an AUC exceeding 0.90 but also of further differentiating between TCMR and ABMR. Furthermore, a simplified model based on six genes exhibited a negative predictive value as high as 90% for excluding ABMR which provides a substantial evidentiary basis for its clinical application [53].
Concurrently, exosomal miRNAs have also demonstrated immense diagnostic potential. Researchers have identified a triad miRNA signature consisting of hsa-miR-21-5p, hsa-miR-31-5p, and hsa-miR-4532 which achieved an AUC of 0.77 for diagnosing AR in a validation cohort and indicates promising application prospects [54]. Notably, the elevation of miR-21 levels which is closely associated with fibrosis pathways may serve as an early warning signal for chronic injury [55].
Other emerging biomarkers
Blood gene expression profiles (Kidney Solid Organ Response Test)
The Kidney Solid Organ Response Test (kSORT) is a risk scoring system based on the expression levels of 17 genes in peripheral blood designed to evaluate the immune status of the recipient by distinguishing between immune quiescence and high-risk rejection [56]. Early data from the AART study indicated that kSORT achieved an AUC as high as 0.94 for predicting AR [57]. However, subsequent independent validation studies have reported more modest performance. For example, a recent study showed that the kSORT assay achieved an AUC of approximately 0.71, indicating only moderate predictive accuracy [58]. These discrepancies suggest that the predictive performance of kSORT may be less robust than initially reported and may vary across different clinical settings and patient populations. The prevailing consensus further indicates that peripheral blood transcriptomic biomarkers are susceptible to systemic nonspecific inflammatory signals, which may limit their organ specificity compared with urinary biomarkers.
Torque teno virus
Torque teno virus (TTV) is a nonpathogenic virus that is highly prevalent within healthy populations [59]. Its uniqueness lies in the fact that viral replication levels are directly influenced by the host’s immune status and are inversely correlated with the intensity of immunosuppression which leads to its recognition as an endogenous immune thermometer. Specifically, a higher TTV load suggests that immunosuppression may be excessive, thereby increasing the risk of infection, whereas a lower load may imply insufficient immunosuppression and a correspondingly elevated risk of rejection [60]. Currently, immune monitoring strategies based on TTV have advanced to the stage of prospective clinical trials such as the TTV-GUIDE IT study which aims to explore the dynamic adjustment of immunosuppressant dosages based on TTV loads to provide a novel pathway for achieving individualized precision immunotherapy [61].
Functional imaging: macroscopic mapping and physical quantification of histopathology
Liquid biopsy and functional imaging collectively constitute a noninvasive dual-pathway monitoring system for the assessment of renal allograft status. The former detects biomarkers within blood or urine at the molecular level whereas the latter provides macroscopic structural and functional information regarding the organ at the tissue level [62]. Conventional anatomical imaging modalities such as ultrasound and computed tomography (CT) primarily offer morphological information concerning the kidney including size, location, collecting system dilation, and the patency of major blood vessels. However, these parameters often lack sufficient specificity for the detection of early and active renal allograft rejection and particularly for the differentiation of its subtypes. Advances in functional imaging technologies have addressed this limitation by quantifying the physical and physiological parameters of the tissue. These technologies are capable of measuring renal elasticity or stiffness, microvascular perfusion, water molecular diffusion capacity, and blood oxygenation levels [63]. By acquiring these quantitative parameters that reflect intrinsic tissue characteristics, functional imaging reveals early pathophysiological changes undetectable by conventional anatomical imaging and provides novel perspectives for the early identification and classification of rejection. Importantly, functional imaging modalities should not be interpreted as standalone diagnostic tools for rejection. Similar to molecular biomarkers such as dd-cfDNA and urinary chemokines, their specificity for rejection remains limited, as many imaging-derived parameters reflect nonspecific tissue alterations including fibrosis, edema, and perfusion changes. Therefore, the primary value of these techniques lies not in directly diagnosing rejection, but in phenotyping the structural and functional status of the allograft at the tissue level.
Ultrasound elastography
Chronic allograft injury manifests primarily as interstitial fibrosis and tubular atrophy. Fibrosis leads to increased stiffness within the renal parenchyma which is a change in physical properties that can be quantified via elastography. Shear wave elastography (SWE) represents the most established elastography technique currently employed in the field of renal transplantation. Its operating principle involves the emission of acoustic radiation force impulses by the probe to generate shear waves within the tissue followed by the measurement of their propagation speed to calculate Young’s modulus which characterizes tissue stiffness. Extensive evidence has demonstrated a significant positive correlation between renal cortical stiffness measured by SWE and the degree of fibrosis confirmed by biopsy, thereby providing a foundation for its clinical application in noninvasive tissue characterization. Clinical studies further support both the diagnostic and prognostic value of SWE. A prospective study correlating SWE with histopathological findings according to the Banff classification showed that renal cortical stiffness increases progressively with the severity of interstitial fibrosis and tubular atrophy, with mean values of 39.86 kPa in patients without fibrosis and higher thresholds identified for distinguishing moderate and severe stages. Receiver operating characteristic analysis further identified cutoff values of 45.09 kPa and 52.06 kPa for differentiating fibrosis stages, indicating its potential for noninvasive grading of chronic injury [64]. In addition to its diagnostic capability, another study demonstrated that medullary stiffness parameters are independently associated with adverse clinical outcomes, and incorporating SWE into conventional clinical models improves prognostic discrimination, highlighting its role in risk stratification after transplantation [65]. Collectively, these findings indicate that SWE not only enables quantitative assessment of renal fibrosis but also provides complementary information for both disease staging and prognostic evaluation. However, it must be noted that algorithmic discrepancies between different devices may result in systematic bias in measurements, so clinical practice should adhere to device-specific reference ranges. Compared with traditional Doppler-derived resistive index (RI), which is susceptible to multiple hemodynamic confounders, SWE offers a more direct and objective quantification of tissue properties. Indeed, comparative studies have shown that SWE achieves higher diagnostic accuracy for chronic allograft injury than RI, underscoring its potential as a reliable noninvasive tool in clinical practice [66].
Contrast-enhanced ultrasound
AR is frequently accompanied by microvascular inflammation and interstitial edema which lead to increased microcirculatory resistance and reduced perfusion. Conventional ultrasound lacks the capability to detect low-velocity blood flow at the capillary level whereas CEUS achieves this objective through the utilization of microbubble contrast agents. CEUS employs microbubbles encapsulated with gases such as sulfur hexafluoride found in products like SonoVue (Bracco Imaging) as contrast agents. The diameter of these microbubbles is comparable to that of red blood cells so they remain strictly confined within the intravascular space. Under the influence of low mechanical index ultrasound waves, the microbubbles generate nonlinear harmonic resonance which renders their echo signals significantly stronger than those of the surrounding tissue. Unlike iodine or gadolinium contrast agents utilized in CT and magnetic resonance imaging (MRI), microbubbles are excreted through the lungs via respiration and do not undergo renal metabolism. This lack of nephrotoxicity makes them an ideal choice for assessing perfusion in patients with impaired renal function [67]. A series of quantitative parameters reflecting renal hemodynamics can be derived by analyzing the time-intensity curves generated by CEUS. Among these parameters, time to peak (TTP) and rise time (RT) allow for the assessment of blood flow velocity and perfusion resistance while the AUC and perfusion index are utilized to quantify local blood volume and comprehensive perfusion status, respectively [68].
Distinguishing AR from delayed graft function (DGF), primarily caused by ATN, remains a clinical challenge due to overlapping hemodynamic patterns. While earlier conceptual models suggested that ATN might preserve microvascular kinetics similar to normal kidneys, clinical evidence reveals complex perfusion alterations in both conditions [69]. Microvascular inflammation and endothelial swelling during AR typically result in reduced peak enhancement and decreased wash-in perfusion index on CEUS [70]. Kinetic parameters such as TTP and RT show some numerical increase in AR kidneys compared to stable allografts, but these differences are not statistically significant, whereas DGF kidneys exhibit the longest TTP and RT values. This overlap indicates that TTP and RT alone may lack sufficient specificity to reliably differentiate these entities. Instead, quantitative assessment of contrast intensity and integrated diagnostic models, including RI and clinical biomarkers, have shown high discriminative performance, with receiver operating characteristic analyses indicating AUC values ranging from 0.72 to 0.83 for key CEUS parameters such as renal artery TTP and RT, as well as TTP in the cortex and medulla [71]. These findings highlight the value of combining multiple quantitative metrics to improve diagnostic accuracy. Nevertheless, the lack of standardized diagnostic thresholds across different imaging platforms limits the broad clinical application of these quantitative CEUS metrics.
Multiparametric magnetic resonance imaging
MRI not only provides high-resolution anatomical images but also employs functional sequences to characterize the motion of water molecules, oxygenation status, and blood perfusion in tissues which is an approach collectively termed mpMRI. Functional MRI and ultrasound technologies translate microscopic pathological changes into quantifiable physical parameters. Table 2 outlines the physical principles and pathophysiological interpretations of these imaging modalities. Crucially, the clinical utility of this multiparametric framework extends beyond snapshot diagnostics to the dynamic monitoring of therapeutic efficacy. By tracking the longitudinal restoration of physiological signatures, such as tissue oxygenation, water diffusion, and microvascular perfusion, mpMRI facilitates an early and noninvasive assessment of treatment response. This capability enables more proactive management of rejection, often preceding the recovery of traditional and delayed functional markers such as sCr [72].
Blood oxygen level-dependent magnetic resonance imaging: monitoring renal hypoxia
The underlying principle of blood oxygen level-dependent MRI (BOLD-MRI) rests upon the paramagnetic properties of deoxyhemoglobin. Deoxyhemoglobin induces local magnetic field inhomogeneities within the tissue which accelerates transverse relaxation and manifests as an increase in the effective transverse relaxation rate known as R2* [73]. Consequently, the R2* value is inversely correlated with tissue oxygen partial pressure where a higher R2* value typically indicates more severe tissue hypoxia [74].
In the context of renal allograft rejection, the application of BOLD-MRI has unveiled a complex phenomenon often described as an oxygenation paradox. While the vasculitis and edema accompanying AR theoretically restrict blood flow and induce tissue hypoxia, clinical studies have observed that renal medullary R2* levels may paradoxically decrease or remain unchanged during the early stages of rejection which suggests a relative improvement in oxygenation. The potential mechanism underlying this phenomenon is that renal oxygen consumption primarily supports the active reabsorption of sodium ions by the renal tubules. Tubulitis occurring in early rejection impairs tubular function and significantly reduces sodium reabsorption along with its requisite oxygen consumption. When the reduction in oxygen consumption exceeds the decline in oxygen supply caused by compromised blood flow, the actual oxygen content within the tissue rises relatively and manifests as a decrease in R2* values on BOLD-MRI [75]. Furthermore, BOLD-MRI provides a noninvasive measure of renal tissue oxygenation. Longitudinal assessment of stable kidney transplant recipients and their donors demonstrated that cortical and medullary R2* values can track changes in renal oxygenation over time, highlighting their potential utility in monitoring allograft physiology [76]. This unique feedback mechanism enables BOLD-MRI to sensitively detect early impairments in tubular function but simultaneously adds complexity to the interpretation of imaging results [77]. Therefore, to evaluate renal oxygenation regulation and reserve function more comprehensively, clinical practice often incorporates the furosemide stress test for integrated analysis.
Diffusion-weighted imaging: detecting inflammatory infiltration
Diffusion-weighted imaging (DWI) relies on the physical principle of applying diffusion-sensitizing gradient fields to measure the Brownian motion of water molecules within tissues and utilizes the apparent diffusion coefficient (ADC) for quantitative analysis. In the clinical context of renal allograft rejection, this technology demonstrates clear diagnostic value. During renal allograft rejection, the infiltration of the renal interstitium by large numbers of inflammatory cells, such as T cells and macrophages, together with accompanying cytotoxic edema, leads to a marked narrowing of the extracellular space. This process severely restricts the free diffusion of water molecules and characteristically manifests as a significant reduction in ADC values [78]. Previous studies have suggested that the diagnostic performance of DWI may be influenced by acquisition parameters such as b-values, with higher b-values potentially improving sensitivity to diffusion restriction [79]. However, according to a systematic review, current evidence indicates that although ADC values are associated with renal dysfunction and histopathological changes such as fibrosis, DWI remains limited in its ability to differentiate among the various underlying causes of allograft dysfunction, including rejection, ATN, and other injuries [80]. Instead, DWI appears to have greater utility in assessing the severity of renal impairment and microstructural alterations, rather than serving as a specific diagnostic tool for rejection.
Arterial spin labeling: contrast-free perfusion quantification
The physical principle underlying arterial spin labeling (ASL) involves the application of radiofrequency pulses to magnetically label water molecules within arterial blood, thereby utilizing them as endogenous tracers. Images are acquired as this labeled blood perfuses the kidney and are subsequently subtracted from control images obtained without labeling which allows for the precise calculation of renal blood flow [81]. This methodology completely obviates the need for exogenous gadolinium-based contrast agents and thereby eliminates the associated risk of nephrogenic systemic fibrosis and ensures a high safety profile. An initial study has demonstrated that renal cortical perfusion measured via ASL is significantly reduced in a patient experiencing AR and that this parameter exhibits a distinct correlation with the eGFR [82]. However, it is important to note that this reduction in perfusion is not inherently specific to rejection, as similar hemodynamic alterations were also observed in cases of allograft artery dissection and hydronephrosis within the same study. Furthermore, comparative research indicates that ASL may be a more promising imaging technique than BOLD-MRI for diagnosing renal allograft injury, as it provides a more sensitive assessment of cortical perfusion deficits [83]. While the potential exists to utilize both modalities to explore the pathophysiological phenomenon of perfusion-oxygenation mismatch, the superior diagnostic accuracy of ASL alone suggests it may be the more robust primary functional tool in this context. However, the feasibility of widespread clinical implementation, particularly in terms of cost-effectiveness and accessibility, requires further evaluation.
Reproducibility and standardization
Despite their promising diagnostic performance, functional imaging techniques are subject to limitations related to reproducibility and standardization. Quantitative parameters derived from modalities such as SWE, CEUS, and mpMRI may vary across imaging equipment, software algorithms, acquisition protocols, and operator-dependent factors, particularly in ultrasound-based techniques. This heterogeneity may limit the comparability of results across studies and reduce the generalizability of proposed diagnostic thresholds. Although efforts toward standardization, including the development of consensus protocols and harmonization of imaging parameters, are ongoing, these approaches have not yet been uniformly implemented in clinical practice. Therefore, imaging findings should be interpreted in conjunction with clinical context and other biomarkers, and further multicenter studies are required to establish robust and reproducible diagnostic criteria.
Artificial intelligence and multimodal fusion: the next-generation diagnostic system
Individual biomarkers or standalone imaging modalities often suffer from insufficient specificity or susceptibility to confounding factors. The future trajectory of diagnostics inevitably lies in the deep integration of multimodal data for which AI and ML provide robust tools. Despite their promise, the clinical implementation of multimodal integration and AI remains constrained by practical challenges, including cost, infrastructure requirements, regulatory considerations, and data governance. In the near term, these approaches are more likely to serve as adjunctive tools within existing workflows, with broader adoption dependent on multicenter validation and standardized data frameworks.
The iBox system
The iBox system represents a milestone in the field of renal transplant prognosis assessment as it stands as the first prognostic scoring system to obtain favorable regulatory qualification opinions from European and American authorities. Its core breakthrough lies in transcending the binary diagnostic paradigm of determining the presence or absence of rejection and instead focusing on the dynamic prediction of the probability of long-term allograft survival. The system constructs a risk model by integrating multidimensional clinical data which includes renal function indices such as eGFR and proteinuria, the immunological profile defined by the specificity and strength of anti-human leukocyte antigen antibodies, and allograft biopsy pathology scores when available [84]. Utilizing Cox proportional hazards models and ML algorithms, the iBox system is capable of precisely quantifying the risk of allograft loss for patients over future horizons of 3, 5, 7, and even 10 years [85]. Currently, iBox has emerged as a pivotal surrogate endpoint for evaluating the efficacy of novel therapies in clinical trials. The predictive performance of iBox has been validated across large international cohorts as well as in specific populations such as pediatric transplant recipients, demonstrating robust discrimination and calibration across different clinical settings [84,86]. Currently, iBox has emerged as a promising surrogate endpoint for evaluating therapeutic efficacy in clinical trials, although further studies are needed to clarify its implementation and real-world applicability.
Banff automation systems and digital pathology
In the domain of pathological diagnosis, AI is profoundly revolutionizing the traditional Banff pathology assessment system. Automated diagnostic systems powered by deep learning are capable of performing pixel-level analysis on whole-slide images of renal biopsies to automatically identify and segment key tissue structures including glomeruli, renal tubules, interstitium, and microvessels while precisely quantifying the infiltration density of inflammatory cells [87]. The technical architecture underlying these systems typically integrates U-Net networks for tissue segmentation and convolutional neural networks for lesion classification while employing multi-instance learning strategies to synthesize information across the entire slide. A distinct advantage is that AI models demonstrate accuracy in diagnosing TCMR and ABMR comparable to that of experienced pathologists and completely eliminate the subjective variability associated with manual interpretation [88]. More importantly, AI is capable of extracting sub-visual features indiscernible to the human eye which have demonstrated superior predictive accuracy compared to traditional Banff scores regarding the risk of allograft loss and therapeutic response [89].
Radiomics
Radiomics is an emerging imaging analysis approach that extracts high-dimensional quantitative features from medical images to characterize tissue heterogeneity and microstructural properties [90]. Although initially developed in oncology, this methodology has recently been extended to nonneoplastic diseases, including organ transplantation. In renal transplantation, ultrasound radiomics has been explored as a noninvasive approach for evaluating posttransplant renal function by extracting quantitative features from two-dimensional ultrasound images and integrating them with clinical or conventional ultrasound variables. In one retrospective study of 233 renal transplant recipients, ML models based on clinical, ultrasound image, and radiomic features achieved moderate discriminatory performance for distinguishing normal from abnormal posttransplant renal function, with test-set AUCs ranging from 0.788 to 0.839, and also showed potential for stratifying the severity of renal functional impairment [91].
However, several limitations should be acknowledged. The current evidence is largely derived from single-center retrospective studies with relatively limited sample sizes, and the generalizability of these models remains to be validated. In addition, variability in image acquisition, feature extraction, and model construction may affect reproducibility across institutions. Future studies with standardized imaging protocols, external validation cohorts, and prospective designs are warranted to establish the robustness and clinical utility of radiomics in kidney transplantation [92].
Conclusion
The landscape of noninvasive diagnosis for renal allograft rejection is currently characterized by a distinct trend toward the multimodal integration of liquid biopsy, functional imaging, and AI. Based on existing evidence, a stratified and progressive diagnostic pathway extending from screening to definitive diagnosis has begun to emerge. The establishment of this stratified diagnostic model is primarily based on the superior diagnostic efficacy demonstrated in clinical validation. Table 3 summarizes representative clinical studies underpinning this field, along with commonly reported diagnostic thresholds for each modality. This paradigm involves preliminary screening utilizing cost-effective biomarkers such as urinary chemokines, followed by high-precision confirmation via dd-cfDNA for patients exhibiting abnormalities. Subsequently, functional imaging is employed to evaluate the nature and localization of the injury, while invasive biopsy is reserved solely for cases where it is strictly necessary. This approach holds the promise of achieving the early identification and precision management of rejection while simultaneously minimizing reliance on invasive procedures.
Nevertheless, the comprehensive transition toward a noninvasive era continues to face critical challenges including the health economic justification for high diagnostic costs and the urgent necessity for standardization across different testing platforms. Looking forward the advancement of point-of-care testing technologies and portable imaging devices combined with cloud-based AI analysis is expected to enable transplant follow-up models to transcend hospital boundaries. This evolution will facilitate patient-centered management characterized by continuous and remote intelligent monitoring. This transformation will ultimately drive posttransplant management from a passive event-driven approach to an active risk-prediction model and thereby establish a new paradigm for enhancing long-term allograft survival.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Print
