Abstract: TH-PO0129
MRI Radiomics-Based Assessment of Mayo Imaging Classification (MIC) and Longitudinal Radiomic Changes in ADPKD
Session Information
- Integrative Genomics and Multi-Omics in Noncystic Genetic Kidney Disease
October 22, 2026 | Location: Exhibit Hall A, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Genetic Diseases of the Kidneys
- 1202 Genetic Diseases of the Kidneys: Non-Cystic (Complex and Non-Cystic Monogenic)
Authors
- Kremer, Linnea E., The University of Chicago Division of the Biological Sciences, Chicago, Illinois, United States
- Chapman, Arlene B., The University of Chicago Division of the Biological Sciences, Chicago, Illinois, United States
- Armato, Sam, The University of Chicago Division of the Biological Sciences, Chicago, Illinois, United States
Background
MIC is a prognostic imaging biomarker for kidney function decline in ADPKD patients; low-MIC (1A/1B) patients maintain relatively preserved kidney function compared with high-MIC (1C/1D/1E) patients. Quantitative features extracted from MRI scans through radiomics analysis were used to (1) classify low- and high-risk MIC patients and (2) explore the clinical relevance of longitudinal feature changes within the non-cystic kidney parenchyma (NCKP) of ADPKD patients.
Methods
The largest baseline cross-sectional T2-weighted MRI coronal image of each kidney was selected from HALTA-PKD studies (low MIC=138, high MIC=324). Corresponding images from 24-, 48-, and 60-month follow-up scans were identified from a subset of low-MIC (n=93) and high-MIC (n=188) patients for feature extraction. Clinical features (age, sex, estimated glomerular filtration rate (eGFR), height-corrected total kidney volume (htTKV)) also were obtained. A convolutional neural network was used to segment the kidney and subsequently the NCKP. Baseline scans were used to identify key radiomic features that discriminate between MIC groups. A multivariate linear regression model using baseline clinical and radiomic features predicted disease progression (%eGFR decline at 60 months). Linear mixed-effects modeling assessed longitudinal feature trajectories among low- and high-MIC groups.
Results
The areas under the receiver operating characteristic curve for differentiation of low- vs. high-MIC patients were 0.80 and 0.90 in the NCKP and entire kidney, respectively. Prediction for %eGFR decline at 60-months yielded improved correlation by combining radiomic and clinical features (r=0.58) vs. clinical features alone (r=0.43). Four radiomic features had significantly different trajectories between MIC groups.
Conclusion
NCKP radiomic features discriminate between low- and high-MIC groups at baseline, predict change in eGFR over 60 months, and change over time differently across MIC classes. NCKP radiomic features may be useful predictive and associative biomarkers in ADPKD.
Funding
- NIDDK Support