Machine Learning with Multiparametric MRI and Clinical Biomarkers for Noninvasive Renal Interstitial Fibrosis Staging

Area of Science:

• Nephrology
• Radiology
• Biomedical Engineering

Background:

• Renal interstitial fibrosis (RIF) assessment traditionally relies on invasive kidney biopsies.
• Developing noninvasive methods for RIF evaluation is crucial for patient management and reducing healthcare burdens.

Purpose of the Study:

• To develop and validate a noninvasive model for identifying severe RIF using multiparametric MRI and clinical biomarkers.
• To compare the performance of this novel model against existing clinical and MRI-only approaches.

Main Methods:

• A prospective study involving 116 patients with biopsy-confirmed renal disease.
• Quantitative MRI parameters were extracted from various sequences (IVIM, ASL, phase-contrast, T1 mapping, BOLD).
• A random forest model was developed integrating multiparametric MRI and clinical data to classify RIF as mild (<25%) or severe (≥25%).

Main Results:

• The integrated random forest model achieved an AUC of 0.89 in the test set, significantly outperforming clinical-only (AUC 0.63), MRI-only (AUC 0.63), and LASSO regression (AUC 0.73) models.
• The model demonstrated high sensitivity (0.91) and specificity (0.73) for identifying severe RIF.
• Superior calibration (Brier score 0.154) and net clinical benefit were observed compared to benchmark models.

Conclusions:

• An integrated MRI-clinical random forest model shows significant promise for the noninvasive identification of severe renal interstitial fibrosis.
• This noninvasive approach could potentially reduce the need for invasive kidney biopsies.
• Further external prospective validation is recommended to confirm the model's generalizability and clinical utility.