Abstract: TH-PO025
Magnetic Resonance Imaging Radiomics Analysis of Noncystic Kidney Parenchyma to Differentiate among Mayo Imaging Classification Classes
Session Information
- Augmented Intelligence for Prediction and Image Analysis
October 24, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Kremer, Linnea E., The University of Chicago, Chicago, Illinois, United States
- Chapman, Arlene B., The University of Chicago Medicine, Chicago, Illinois, United States
- Armato, Sam, The University of Chicago, Chicago, Illinois, United States
Background
Mayo Imaging Classification (MIC) serves as a prognostic biomarker for the prediction of kidney function decline of patients with autosomal dominant polycystic kidney disease (ADPKD). Radiomics analysis—the extraction of quantitative image features—may provide additional power in identifying patients that are fast progressors to kidney function decline. Specifically, characterizing the non-cystic kidney tissue using radiomic features extracted from magnetic resonance images (MRI) may differentiate among MIC classes. This work (1) determined whether radiomic features can differentiate between low/intermediate- and high-risk MIC classes in the non-cystic components of the kidney and (2) investigated the effect of image pre-processing on subsequent classification.
Methods
Radiomic features were extracted from T2-weighted fat saturation MRI scans of 138 MIC 1A/1B patients and 324 MIC 1C/1D/1E patients for classification. A pre-processing pipeline was developed for (1) normalization of MRI signal intensity (z-score or psoas muscle as a reference tissue), (2) pixel resampling (upsampling or downsampling) schemes, and (3) gray-level discretization (fixed bin size method with 8-256 gray levels for discretization). Area under the receiver operating characteristic curve (AUC) was used as a performance-assessment metric.
Results
The range of AUC values across pre-processing parameters was 0.69-0.85. Radiomic features for classification include gray-level co-occurrence matrix inverse difference and first-order kurtosis.
Conclusion
The results indicate the potential of radiomics to distinguish between patients based on low/intermediate- and high-risk MIC classification.
Area under the receiver operating characteristic curve (AUC) values in classifying Mayo Imaging Classification using radiomic features extracted from the non-cystic kidney parenchyma using fixed bin size discretization. The dotted line at an AUC of 0.5 is random guessing.
Funding
- NIDDK Support