• ABSTRACT
    • Total hip arthroplasty (THA) is the most successful treatment for end-stage joint degeneration; however, implant loosening remains a common reason for revision and its detection on radiography can be challenging. Three-dimensional multi-spectral magnetic resonance imaging (3D-MSI-MRI) enables imaging near metal implants and combining this with radiomics may enhance detection of loosening. In this cohort study, 3D MSI-MRI acquisitions were obtained from 76 patients with THA, including 36 symptomatic patients with confirmed femoral component loosening and 40 asymptomatic patients. Periprosthetic trabecular bone of the femoral stem was segmented and divided into Gruen zones. Two-dimensional overlapping patches (12 × 12 mm2 to 64 × 64 mm2) were extracted from the four proximal Gruen zones and combined full proximal region. A total of 96 radiomic features were computed per patch, and dimensionality was reduced using principal component analysis. Logistic regression with five-fold cross-validation was used to classify loosening. Zonal radiomic models outperformed the full proximal region models, with Gruen Zones 1 (lateral), 7 (medial), and 14 (posterior) achieving high predictive performance (AUC > 0.79) at larger patch sizes. Feature reduction consistently retained 11-17 principal components across regions, suggesting redundancy in radiomic data. Larger patches improved classification but were limited by anatomical constraints in zones with less trabecular bone. Zonal radiomic differences likely reflect stress-induced trabecular remodeling near the implant. MSI-MRI-based zonal radiomics can detect femoral implant loosening with high accuracy. These results support the future development of radiomic-based decision support tools to aid in early loosening detection with the potential to improve patient outcomes. STATEMENT OF CLINICAL SIGNIFICANCE: This study demonstrates that MSI-MRI based radiomic features can classify the qualitative diagnosis of implant loosening in patients with THA. The findings provide a foundation for future radiomic tools to assist in early, quantitative detection of aseptic loosening during routine clinical imaging.