Identification of Knee Osteoarthritis Based on Bayesian Network: Pilot Study.

Area of Science:

• Biomedical informatics
• Machine learning in healthcare
• Osteoarthritis research

Background:

• Early identification of knee osteoarthritis (OA) is crucial for effective treatment and cost reduction.
• Existing classification models for OA face limitations in data processing and complexity, hindering clinical application.

Purpose of the Study:

• To develop a Bayesian network (BN)-based classification model for identifying individuals with knee OA.
• To establish a prescreening tool that aids healthcare professionals in decision-making for knee OA.

Main Methods:

• A 3-level BN structure was employed, optimized using the Bayesian Search (BS) learning algorithm.
• Model parameters were determined via the expectation-maximization algorithm.
• Performance was assessed using accuracy, AUC, specificity, sensitivity, PPV, and NPV, with comparisons to other models and inclusion of physical fitness tests.

Main Results:

• The proposed BN model achieved higher or equal performance metrics compared to other classification models (Accuracy: .754, AUC: .78, Specificity: .78, Sensitivity: .73).
• Significant improvements were observed against traditional BN models (e.g., 6.3% increase in accuracy, 4.0% in AUC).
• Incorporating physical fitness tests substantially enhanced model performance, notably increasing accuracy by 10.6% and AUC by 16.4%.

Conclusions:

• The developed BN model demonstrates significant promise for knee OA classification compared to existing methods.
• This model can be practically implemented as a prescreening tool in clinical settings.
• Adoption of this tool can enhance healthcare quality for the elderly and reduce overall medical expenditures.