Navigating the diagnostic 'gray zone': prospective evaluation of an integrated MRI-Biomarker model for renal allograft triage.

BACKGROUND

A critical dilemma in post-transplant surveillance is determining when minor functional changes (subtle serum creatinine fluctuations or slow renal function decline) or ambiguous ultrasound findings (including mildly elevated resistive index or cortical echogenicity) warrant invasive biopsy and treatment modification. We aimed to develop a non-invasive integrated tool combining advanced diffusion MRI with functional biomarkers to stratify patients within this diagnostic gray zone.

MATERIALS AND

METHODS

In this prospective study, 120 renal transplant recipients underwent multi-b-value diffusion-weighted imaging at 3.0 T within 48 h prior to an allograft biopsy. Quantitative parameters of six diffusion MRI models were extracted from renal cortex and medulla. Clinical indicators, including serum HCO3-, were recorded. To establish a clear endpoint, patients were categorized into a Treatment-Changed group (presence of actionable active pathology, requiring modified therapies) and a Treatment-Unchanged group (normal or solely chronic/non-reversible lesions). Multivariable logistic regression identified independent predictors. Diagnostic performance was assessed using area under the curve (AUC) using repeated stratified 5-fold cross validation with 10 repeats and bootstrap resampling. Kaplan-Meier analysis assessed the association between model-based risk and subsequent graft functional decline, defined as a 30% or greater decrease in estimated glomerular filtration rate (eGFR) from baseline.

RESULTS

The cortical distributed diffusion coefficient (DDC) and serum HCO3-, emerged as independent predictors of treatment-changed group. The integrated model achieved an AUC of 0.941, significantly outperforming the standard clinical reference model (eGFR, proteinuria, and ultrasound; AUC 0.563). Net reclassification improvement (0.479) demonstrated the model correctly reclassified 45.6% of patients. Furthermore, model-derived high-risk status significantly predicted long-term graft functional decline (log-rank p = 0.005).

CONCLUSION

This integrated multiparametric protocol provides promising preliminary decision support for renal allograft surveillance. By noninvasively identifying the potential presence of actionable active pathology, this preliminary model may help optimize biopsy referrals and stratify patients who potentially require treatment modification. However, given the single-centre design, external validation is required before clinical implementation.