Comparison of a quantitative MRI-based radiomics marker with semiquantitative signal alteration scores in the infrapatellar fat pad: Data from the Osteoarthritis Initiative.
Authors
Affiliations (10)
Affiliations (10)
- Department of Radiology, Balgrist University Hospital, Zurich, Switzerland; Faculty of Medicine, University of Zurich, Zurich, Switzerland. Electronic address: [email protected].
- Rheumazentrum Herne, Ruhr University Bochum, Herne, Germany. Electronic address: [email protected].
- Department of Radiology and Biomedical Imaging, University of California, San Francisco, USA. Electronic address: [email protected].
- Department of Radiology and Biomedical Imaging, University of California, San Francisco, USA. Electronic address: [email protected].
- Department of Radiology and Biomedical Imaging, University of California, San Francisco, USA. Electronic address: [email protected].
- Department of Epidemiology and Biostatistics, University of California, San Francisco, USA. Electronic address: [email protected].
- Department of Internal Medicine, U.C. Davis Health, Sacramento, California, USA. Electronic address: [email protected].
- Department of Epidemiology and Biostatistics, University of California, San Francisco, USA. Electronic address: [email protected].
- Department of Epidemiology and Biostatistics, University of California, San Francisco, USA. Electronic address: [email protected].
- Department of Radiology and Biomedical Imaging, University of California, San Francisco, USA. Electronic address: [email protected].
Abstract
To develop a quantitative MRI-based radiomics marker of the infrapatellar fat pad (qIPFP) based on semiquantitative IPFP signal alteration scores (sqIPFP), and to evaluate whether its associations with symptoms and imaging features of knee osteoarthritis are comparable to those of sqIPFP. This study included 2640 participants from the Osteoarthritis Initiative, split into training (n = 2112) and test (n = 528) sets. Deep learning-based IPFP segmentation was performed on baseline intermediate-weighted fat-suppressed 3.0-Tesla MRI. Stability selection with bootstrap-resampled ElasticNet regression (≥80% selection frequency) identified radiomic features for qIPFP trained on a composite score combining sqIPFP size and intensity. Linear regression models, including age, sex, race, and body mass index as covariates, were used to compare qIPFP against the sqIPFP composite, size, and intensity scores for cross-sectional associations with the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) pain score and Whole-Organ Magnetic Resonance Imaging Score (WORMS) scores of cartilage lesions, meniscal lesions, and bone marrow lesions (BML). Forty stable features were included in the qIPFP marker, which showed adjusted R² values of 0.15 (95% CI 0.10-0.23) for WOMAC pain, 0.13 (95% CI 0.09-0.20) for cartilage lesions, 0.10 (95% CI 0.05-0.17) for meniscal lesions, and 0.08 (95% CI 0.05-0.14) for BML. There was no evidence of a difference compared to sqIPFP scores (all p ≥ 0.09), except for BML, where the sqIPFP composite (0.12, 95% CI 0.08-0.19; p = 0.042) and intensity scores (0.12, 95% CI 0.08-0.19; p = 0.038) yielded higher adjusted R². A quantitative radiomics marker of infrapatellar fat pad signal alterations showed cross-sectional associations with knee osteoarthritis symptoms and imaging features comparable to those of semiquantitative scoring except for bone marrow lesions, with the advantage of fully automated, reader-independent assessment.