Periprosthetic Joint Infection Prediction via Machine Learning: Comprehensible Personalized Decision Support for...

Area of Science:

• Orthopedic Surgery
• Medical Diagnostics
• Artificial Intelligence in Medicine

Background:

• The International Consensus Meeting (ICM) 2018 criteria are widely used for diagnosing periprosthetic joint infection (PJI).
• These criteria are prespecified and fixed, potentially limiting adaptability in diagnosis.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) system for PJI diagnosis.
• To compare the diagnostic performance of the ML system against the established ICM scoring system.

Main Methods:

• An ensemble meta-learner combining 5 algorithms was designed for optimal performance.
• An explanation generator was developed to enhance ML model interpretability.
• Stratified 5-fold cross-validation was performed on a cohort of 323 patients.

Main Results:

• The ML system exhibited superior predictive performance across multiple metrics (accuracy, precision, recall, F1, MCC, AUC).
• ML identified personalized diagnostic features not captured by ICM.
• ML provided interpretable decision support for individual PJI diagnoses.

Conclusions:

• Machine learning is a feasible and competitive tool for PJI diagnosis compared to ICM criteria.
• Adaptive ML models can complement the ICM system for improved PJI diagnosis.
• ML offers a more dynamic approach by constructing diagnostic models from patient data.