Radiomics and Machine Learning Can Differentiate Transient Osteoporosis from Avascular Necrosis of the Hip

• DOI: 10.3390/diagnostics11091686
• Publication Date: 2021-09-16T01:47:11Z
• Authors: Michail E. Klontzas; Georgios C. Manikis; Katerina Nikiforaki; Evangelia E. Vass...; Konstantinos Span...; Ioannis Stathis; George A. Kakkos; Nikolas Matthaiou; Aristeidis H. Zibis; Kostas Marias; Apostolos H. Kara...
• Keywords: Avascular Necrosis

Differentiation between transient osteoporosis (TOH) and avascular necrosis (AVN) of the hip is a longstanding challenge in musculoskeletal radiology. The purpose this study was to utilize MRI-based radiomics machine learning (ML) for accurate differentiation two entities. A total 109 hips with TOH 104 AVN were retrospectively included. Femoral heads necks segmented features extracted. Three ML classifiers (XGboost, CatBoost SVM) using 38 relevant trained on 70% validated 30% dataset. performance compared radiologists, general radiologist radiology residents. XGboost achieved best an area under curve (AUC) 93.7% (95% CI from 87.7 99.8%) among models. MSK radiologists AUC 90.6% 86.7% 94.5%) 88.3% 84% 92.7%), respectively, similar 84.5% 80% 89%), significantly lower than (p = 0.017). In conclusion, radiomics-based higher differentiating AVN.