Comparison of Resampling Methods and Radiomic Machine Learning Classifiers for Predicting Bone Quality Using...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Osteoporosis Research

Background:

• Dual X-ray Absorptiometry (DXA) is crucial for bone health assessment.
• Detecting degraded bone structures in DXA images remains a challenge.
• Current quality assessment tools are limited, necessitating advanced analytical methods.

Purpose of the Study:

• To develop and validate a novel approach for detecting degraded bone structures in DXA images.
• To enhance the accuracy and accessibility of bone quality assessment tools for clinicians.
• To improve the early diagnosis and management of bone-related conditions.

Main Methods:

• A dataset of 1531 spine DXA images was analyzed.
• Radiomic features were extracted using Pyradiomics.
• Machine learning classifiers (Logistic Regression, SVM, XGBoost) were trained and evaluated after data resampling (SMOTEENN).

Main Results:

• The Support Vector Machine (SVM) classifier demonstrated superior performance.
• An F-score of 97.5% was achieved using specific radiomic features (GLDM, GLRLM) and SMOTEENN resampling.
• Effective class imbalance handling was crucial for optimal model performance.

Conclusions:

• Radiomic texture features, resampling techniques, and machine learning show significant potential for classifying bone health in DXA images.
• This approach can lead to improved clinical diagnosis and treatment strategies for bone degradation.
• The developed method offers a promising tool for accessible and accurate bone quality assessment.