Explainable Supervised Learning Reveals Radiomics Markers for Vertebral Fracture Detection

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Vertebral fractures (VFs) are common osteoporosis complications, leading to mortality, reduced quality of life, and high healthcare costs.
• Current methods like bone mineral density (BMD) and dual-energy X-ray absorptiometry (DXA) have limitations in predicting fracture risk.

Purpose of the Study:

• To evaluate the predictive accuracy of radiomics markers from lumbar spine CT images for fragility fractures.
• To identify key radiomic features and vertebrae associated with fracture susceptibility.

Main Methods:

• Utilized computed tomography (CT) images of the lumbar spine from subjects with fragility fractures.
• Applied a range of machine learning algorithms (Random Forest, XGBoost, SVM, etc.) with optimized hyperparameters (Grid Search).
• Extracted and analyzed radiomics markers to identify discriminative features.

Main Results:

• Identified specific radiomic markers, including Gray Level Emphasis, Kurtosis, and Zone Entropy, as highly predictive.
• Highlighted the L5 vertebra as particularly relevant for fracture risk assessment.
• Demonstrated that radiomics can capture subtle structural variations indicative of bone quality.

Conclusions:

• Radiomics markers show potential for a more comprehensive bone quality assessment than traditional methods.
• AI-driven analysis of medical images can enable non-invasive, early fracture risk assessment.
• Findings support the development of personalized prevention strategies for osteoporosis-related fractures.