Abstract: SA-PO0059
Prediction of Total Kidney Volume (TKV) Progression in ADPKD Using Radiomics
Session Information
- Artificial Intelligence: Kidney Imaging, Pathology, and Precision Phenotyping
October 24, 2026 | Location: Exhibit Hall A, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Yousef Kalafi, Elham, Perceptive Informatics UK Limited, Nottingham, England, United Kingdom
- Kadumberi, Anand Preshob, Perceptive Informatics UK Limited, Nottingham, England, United Kingdom
- Michalski, Andrew S., Perceptive Informatics UK Limited, Nottingham, England, United Kingdom
- Sood, Rohit, Perceptive Informatics UK Limited, Nottingham, England, United Kingdom
Background
ADPKD is a disorder characterized by progressive renal cyst growth and variable rates of disease progression. Radiomics offers a quantitative imaging approach that can extract high-dimensional features from MRI. This study aims to evaluate whether radiomics can detect renal cyst heterogeneity—invisible to the naked eye—as a tool for predicting disease progression.
Methods
We retrospectively analyzed T2w MRI scans of ADPKD patients (n=25; CRISP study). All images were reviewed by one expert reader (15 years experience) and representative renal cysts were manually segmented on a standardized slice. Three cysts per patient were assessed across baseline (BL) and follow-up (FU). 788 radiomic features were extracted using PyRadiomics from original and wavelet-filtered images, then ranked using Random Forest (RF) feature importance. The top 10 features were used to train an XGBoost regressor with 5-fold cross-validation to predict FU TKV from BL radiomics features.
Results
The model demonstrated predictive performance with a mean absolute error (MAE) of 732cc (+/- 265.3), RMS error of 915cc (+/- 324.5) and, MAE(%) of 43.98% across cross-validation folds. Predicted follow-up TKV showed a positive correlation with observed values, with most points aligned along the identity line (Fig. 1). Performance appeared stronger in patients with less advanced disease, whereas prediction was less robust in patients with very large kidneys or advanced ADPKD, suggesting increased biological heterogeneity and growth variability in later disease stages
Conclusion
BL MRI radiomics may provide a noninvasive approach for predicting progression in ADPKD. These findings suggest that radiomic heterogeneity at baseline captures meaningful prognostic information. Larger multicenter cohorts are needed to validate the model and determine its clinical utility for longitudinal risk stratification.
Funding
- NIDDK Support