Explainable Supervised Learning Reveals Radiomics Markers for Vertebral Fracture Detection

Abstract

Vertebral fractures (VFs) are the most common osteoporosis-related fractures, associated with increased mortality, impaired quality of life, disability, and significant economic health burden. In this study, we evaluated the predictive accuracy of radiomics markers extracted from computed tomography (CT) images of the lumbar spine in subjects with fragility fractures. We deployed a comprehensive suite of state-of-the-art machine learning algorithms: Random Forest, XGBoost, Support Vector Machines (SVM), Decision Tree, Naive Bayes, Logistic Regression, AdaBoost, KNN, with hyperparameters optimized via Grid Search. We identified highly discriminative radiomic markers that provide clinically valuable insights into the radiomic features most relevant to fracture susceptibility. In particular, features such as Gray Level Emphasis, Kurtosis, and Zone Entropy, along with the prominent role of the L5 vertebra, suggest that radiomic marker can capture subtle structural variations that precede fractures and offer a more comprehensive assessment of bone quality compared to conventional bone mineral density and dual-energy X-ray absorptiometry measurements. These findings pave the way for a shift toward AI-driven, non-invasive fracture risk assessment, potentially improving early diagnosis and personalized prevention strategies.Clinical relevance- This work revealed medical image derived biomarkers that can potentially improve early diagnosis and personalized prevention strategies.

Ähnliche Arbeiten