Bayesian Optimization in Restricted Boltzmann Machines for Heart Failure severity estimation.

Area of Science:

• Cardiology
• Machine Learning
• Artificial Intelligence

Background:

• Heart Failure (HF) management is challenged by subjective patient severity evaluation.
• Accurate classification of HF severity (NYHA class) is crucial for effective patient care and reducing hospitalizations.

Purpose of the Study:

• To develop and evaluate a hybrid Restricted Boltzmann Machine (RBM) model for automated Heart Failure severity estimation.
• To classify patients into four New York Heart Association (NYHA) classes using a hybrid RBM classifier.

Main Methods:

• Utilized a hybrid Restricted Boltzmann Machine (RBM) combining generative and discriminative configurations.
• Employed Bayesian optimization for hyper-parameter tuning within a 4-dimensional space.
• Evaluated the model using 5-fold cross-validation on a dataset of 134 patients from three clinical centers.

Main Results:

• Achieved high classification performance: 87.23% accuracy, 87.69% precision, 87.47% recall, and 87.51% f-score.
• Demonstrated the efficacy of the hybrid RBM for HF severity estimation.
• Validated the model on a multi-center patient dataset.

Conclusions:

• Automated HF patient classification using the proposed hybrid RBM model offers a valuable tool for clinical management.
• This approach can overcome subjective assessments and potentially reduce HF-related hospitalizations.
• The model shows promising results for objective and accurate Heart Failure severity stratification.