Radiograph-Based Radiomics Successfully Classify MRI-Based Identification of Femoral Loosening in Patients With Total Hip Arthroplasty-A Preliminary Study.

Abstract

Aseptic loosening is a leading cause of total hip arthroplasty (THA) revision. While three dimensional (3D) multi-spectral magnetic resonance imaging (MSI-MRI) offers high sensitivity for detecting femoral stem loosening, it is less commonly acquired at routine follow-up. This study investigates whether radiograph-based radiomics can detect loosening confirmed by MSI-MRI, improving routine evaluation of implant integration. In this cohort study, 3D MSI-MRI acquisitions and standard AP pelvis radiographs were obtained from 38 symptomatic subjects with confirmed femoral component loosening and 40 asymptomatic subjects. Loosening was identified through qualitative evaluation of MSI-MRI. A 2.5 mm periprosthetic region was segmented from radiographs and divided into six Gruen zones. Radiographs were subdivided into overlapping patches (10 × 10 to 150 × 150 mm²) and a total of 96 radiomic features were extracted per patch. Features were standardized and reduced using principal component analysis. Logistic regression with L2 regularization was trained using five-fold cross-validation to classify symptomatic versus asymptomatic cases. Sixty-four models were trained across zones and patch sizes. Gruen zones 3 (distal, lateral) and 7 (proximal, medial) achieved the highest performance (AUC = 0.76). The full periprosthetic model reached an AUC of 0.74 at 130 × 130 mm². Fewer principal components were required as patch size increased. Radiograph-based radiomics can detect MSI-MRI confirmed femoral component loosening in subjects with THA with high accuracy. As radiographs are routinely acquired in clinical practice, this method offers a scalable, low-cost adjunct to standard post-operative evaluation.