A New Ensemble Classification System For Fracture Zone Prediction Using Imbalanced Micro-CT Bone Morphometrical Data.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Orthopedics

Background:

• Trabecular bone fractures pose a significant health challenge, with current bone mineral density (BMD) prediction methods exhibiting up to 40% error.
• Microstructure-based fracture-zone prediction is an emerging alternative for enhanced fracture risk assessment.

Purpose of the Study:

• To develop and evaluate an automated classification system (CS) for fracture-zone prediction using Ensemble of Imbalanced Learning (EIL) methods.
• To address the challenge of imbalanced datasets where fractured bone area is significantly smaller than intact bone.

Main Methods:

• The proposed CS divides samples into Volumes of Interest (VOIs) and extracts 29 morphometrical parameters as input features.
• It combines imbalanced learning techniques (Random Undersampling, Synthetic Minority Oversampling) with classification algorithms (Multilayer Perceptrons, Support Vector Machines).
• The best performing combination was compared against established EIL methods: RUSBoost, UnderBagging, and SMOTEBagging.

Main Results:

• The proposed CS demonstrated superior performance over existing methods, achieving over 90% in G-Mean and Area Under Curve metrics.
• Sequential forward floating selection identified significant biomechanical parameters, suggesting potential biomarkers for fracture-zone prediction.

Conclusions:

• The developed classification system offers a more accurate method for predicting trabecular bone fractures.
• This approach holds promise for improving fracture risk assessment and identifying novel biomarkers.