Correlation enhanced distribution adaptation for prediction of fall risk

Area of Science:

• Gerontology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Elderly fall risk is a growing concern, necessitating early diagnosis tools.
• Limited labeled data in new patient cohorts hinders accurate diagnosis.
• Heterogeneous health data sources (imaging, wearables) offer proactive monitoring potential.

Purpose of the Study:

• To develop a robust fall-risk prediction model for the elderly using domain adaptation.
• To address challenges in integrating multi-source, multi-domain health data with varying label availability.
• To overcome limitations of traditional machine learning models in handling data distribution shifts.

Main Methods:

• Developed an unsupervised domain adaptation (DA) model to align source and target domains.
• Created a domain-invariant feature representation for enhanced data integration.
• Built a fall-risk prediction model utilizing the aligned feature representations.

Main Results:

• The proposed unsupervised DA approach successfully aligned heterogeneous data domains.
• The developed fall-risk prediction model demonstrated superior performance compared to existing methods.
• Simulation studies and real-world applications validated the model's robustness and accuracy.

Conclusions:

• Domain adaptation techniques are effective in mitigating data discrepancies for improved elderly health monitoring.
• The novel DA-based fall-risk prediction model offers a promising solution for proactive fall prevention in older adults.
• This approach enhances diagnostic accuracy, reducing misdiagnosing risks in new patient cohorts.