Objective

To evaluate the diagnostic performance of combined magnetic resonance imaging (MRI) and ultrasound (US) dual-modality imaging for acute immune rejection (AIR) after kidney transplantation, and to develop and validate a noninvasive predictive model for early detection.

Methods
In this prospective two-center study, 109 kidney transplant recipients (January 2022–June 2024) underwent multiparametric MRI sequences (T1WI, T2WI, DWI, IVIM, DTI) and functional US (B-mode, shear wave elastography) within one week before suspected AIR. Renal biopsy served as the reference standard. Automated graft segmentation was performed using a MATLAB-based deep learning algorithm, followed by radiomics feature extraction via Pyradiomics. Features were selected using correlation analysis, least absolute shrinkage and selection operator (LASSO), and multivariable logistic regression. Three models—MRI, US, and dual-modality—were constructed and visualized with a nomogram. Performance was assessed using receiver operating characteristic (ROC) curves, DeLong tests, and calibration analysis.

Results
Of the 109 patients, 37 had biopsy-confirmed AIR. The MRI model achieved a sensitivity of 84.2% and an area under the ROC curve (AUC) of 0.835; the US model yielded 76.5% sensitivity and an AUC of 0.810. The dual-modality model demonstrated superior diagnostic accuracy, with 87.5% sensitivity and an AUC of 0.905 (P < 0.05 vs. single-modality models). The final nomogram incorporated four independent predictors—perfusion fraction (f), MRI-radscore, renal stiffness, and US-radscore—and showed excellent agreement between predicted and observed outcomes.

Conclusion
The proposed dual-modality MRI-US model provides a reliable, noninvasive tool for early detection of AIR in kidney transplant recipients. Its high diagnostic accuracy and robust calibration support its potential role in guiding personalized clinical management and therapeutic decision-making.