Diagnostic Performance of MRI for Component Loosening in Total Knee Arthroplasty Compared with Radiography

Background Because loosening of total knee arthroplasty (TKA) occurs due to poor osseous integration at component-bone interfaces, interface assessment may be helpful in diagnosing loosening at MRI. Purpose To determine interreader reproducibility for characterizing component interfaces and diagnosing loosening and to evaluate the diagnostic performance of MRI for diagnosing loosening after TKA compared with radiography. Materials and Methods Consecutive knees with TKA that underwent revision between July 2018 and June 2019 and were imaged at MRI and radiography were included in this retrospective study. Interface type (normal, fibrous membrane, fluid, or osteolysis), percent integration (<33%, 33%-66%, or >66%), and presence of bone marrow edema pattern were assessed. Loosening was diagnosed at MRI if no or almost no normal interface was present. Sensitivity and specificity were compared with radiographs by using surgical findings as reference. Gwet agreement coefficient evaluated interreader reproducibility between two readers and multivariable logistic regression assessed risk factors for loosening. Results Among 116 knees in 114 patients (mean age, 63 years ± 10 [SD]; 59 women), 61 of 116 knees (52.6%) had at least one loose component. Interreader reproducibility of MRI was substantial to excellent (Gwet agreement coefficient, 0.67-0.96). Loosening was associated with fluid interface (odds ratio [OR], 20.1; 95% CI: 5.7, 70.9) or osteolysis (OR, 3.1; 95% CI: 1.8, 5.3), absence of any normal interface (OR, 11.8; 95% CI: 6.3, 22.2), poor (<33%) osseous integration (OR, 20.4; 95% CI: 9.7, 42.6), and bone marrow edema pattern (OR, 4.7; 95% CI: 2.8, 7.8). Sensitivity and specificity of MRI for loosening were 84% (27 of 32; 95% CI: 72, 97) and 85% (71 of 84; 95% CI: 77, 92) for the patellar, 31% (eight of 26; 95% CI: 13, 49) and 100% (90 of 90; 95% CI: 100, 100) for the femoral, and 81% (22 of 27; 95% CI: 66, 96) and 98% (87 of 89; 95% CI: 95, 100) for the tibial component, respectively. MRI had higher sensitivity (84% vs 31%; P < .001) but lower specificity (85% vs 96%; P = .003) for patellar component loosening than did radiography, respectively, whereas no evidence of a difference was found for femoral (sensitivity and specificity, MRI vs radiography: 31% vs 46% [P = .20] and 100% vs 99% [P > .99], respectively) or tibial (sensitivity and specificity, MRI vs radiography: 81% vs 70% [P = .16] and 98% vs 97% [P = .32], respectively) component loosening. Conclusion MRI demonstrated substantial to excellent interreader reproducibility and higher sensitivity than did radiography for diagnosing patellar component loosening after total knee arthroplasty. © RSNA, 2022 Online supplemental material is available for this article.