Interpretable Deep-Learning Approaches for Osteoporosis Risk Screening and Individualized Feature Analysis Using...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Osteoporosis Research

Background:

• Osteoporosis requires early detection for effective management.
• Current screening tools and machine learning (ML) models have limitations in accuracy and individualized explanation.
• Existing methods often focus on limited risk factors.

Purpose of the Study:

• To develop an interpretable deep-learning (DL) model for osteoporosis risk screening using clinical features.
• To provide clinical interpretation and individual explanations of feature contributions via explainable artificial intelligence (XAI).

Main Methods:

• Utilized National Health and Nutrition Examination Survey (NHANES) and Korean National Health and Nutrition Examination Survey (KNHANES) data.
• Classified participants based on bone mineral density T-scores.
• Trained a DL model and employed Local Interpretable Model-agnostic Explanations (LIME) to identify significant risk factors and their contributions.

Main Results:

• The DL model achieved high AUC values: 0.851 (femoral neck) and 0.922 (total femur) for NHANES, and 0.827 and 0.912 for KNHANES.
• LIME identified significant risk factors and provided integrated contributions for individual risk interpretation.
• The DL model demonstrated superior performance compared to conventional ML models and clinical tools.

Conclusions:

• The developed DL model significantly outperforms existing osteoporosis screening methods.
• The XAI component facilitates interpretation of individual risk by highlighting feature contributions.
• This interpretable model represents an advancement in osteoporosis risk screening technology.