Machine Learning-Based Automatic Diagnosis of Osteoporosis Using Bone Mineral Density Measurements

Area of Science:

• Biomedical Engineering
• Data Science
• Medical Diagnostics

Background:

• Osteoporosis and osteopenia are common bone diseases in postmenopausal women, increasing fracture risk.
• Dual-energy X-ray absorptiometry (DXA) is the standard diagnostic tool but has limitations in accessibility and fracture prediction.
• Machine learning (ML) offers potential for automated, accurate diagnosis by integrating bone mineral density (BMD) and clinical data.

Purpose of the Study:

• To evaluate the efficacy of various supervised ML algorithms for classifying osteoporosis and osteopenia.
• To assess the impact of different feature selection techniques on ML model performance for BMD data.
• To identify optimal ML models and feature selection methods for automated bone disease diagnosis.

Main Methods:

• Retrospective analysis of BMD data from 142 postmenopausal women.
• Application of supervised ML algorithms: SVM, k-NN, DT, NB, LDA, ANN.
• Utilized feature selection methods: ANOVA, CHI2, MRMR, Kruskal-Wallis, evaluated with 10-fold cross-validation.

Main Results:

• Support Vector Machines (SVM) with ANOVA-selected features achieved the highest accuracy (94.30%) and 100% True Positive Rate for the normal class.
• Statistically selected features generally outperformed traditional BMD regions (e.g., L1-L4, femoral neck).
• CHI2 and MRMR feature selection methods showed robust results, particularly with SVM and k-NN classifiers.

Conclusions:

• ML algorithms combined with data-driven feature selection provide a powerful framework for automated osteoporosis and osteopenia classification.
• ANOVA proved to be the most effective feature selection method, enhancing diagnostic accuracy across classifiers.
• Findings support integrating ML decision support tools for early diagnosis and personalized treatment planning in clinical practice.