A Machine Learning Approach to Microcalorimetric Pattern Classification of Pathogens in Synovial Fluid.

Area of Science:

• Biomedical Engineering
• Infectious Disease Diagnostics
• Computational Biology

Background:

• Periprosthetic joint infection (PJI) diagnosis relies on conventional microbial cultures, which are time-consuming.
• Isothermal microcalorimetry (IMC) offers real-time pathogen growth monitoring but lacks pathogen identification capabilities.
• Developing rapid and accurate PJI diagnostic tools is crucial for timely treatment and improved patient outcomes.

Purpose of the Study:

• To implement and evaluate machine learning (ML) and transfer learning convolutional neural network (CNN) models for detecting and identifying bacterial pathogens in PJI using IMC data.
• To assess the feasibility of distinguishing between aseptic and infected joint fluid samples.
• To determine the accuracy of ML models in identifying specific bacterial strains responsible for PJI.

Main Methods:

• Collected IMC data from 174 aseptic and 239 PJI synovial fluid samples, including five distinct bacterial strains.
• Applied various ML algorithms (XGBoost, multi-layer perceptron, support vector machine, random forest) and three transfer learning CNN models.
• Trained and tested models for binary PJI detection and multiclass pathogen identification.

Main Results:

• The XGBoost binary classifier achieved 100% accuracy in PJI detection.
• Multiclass XGBoost and combined transfer learning CNN models reached 90.3% and 91.5% accuracy in identifying bacterial strains, respectively.
• XGBoost model demonstrated interpretable features, aiding clinical application; specific challenges noted for *Pseudomonas aeruginosa* (PA) recall and *Staphylococcus epidermidis* (SE) precision.

Conclusions:

• Machine learning models can effectively detect and identify PJI pathogens using IMC growth patterns.
• This integration enhances IMC's diagnostic utility, enabling faster PJI diagnosis and targeted antibiotic selection.
• The study validates a novel, rapid approach for PJI diagnostics, potentially improving patient management.