Evolving a Bayesian Classifier for ECG-based Age Classification in Medical Applications.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Cardiology

Background:

• Accurate patient age classification is crucial for medical diagnosis and treatment.
• Electrocardiograms (ECGs) contain rich physiological data that can be leveraged for patient profiling.
• Traditional Bayesian classifiers have limitations in handling complex dependencies in signal data.

Purpose of the Study:

• To classify patients by age using features extracted from ECG signals.
• To develop and compare the performance of Bayesian classifiers, specifically focusing on a genetically evolved network.
• To assess the efficacy of evolutionary computing in discovering Bayesian network structures.

Main Methods:

• Extracted statistical features from ECG signals.
• Converted continuous signal features to a discrete symbolic form via thresholding to reduce dimensionality.
• Developed and compared two Bayesian network discovery methods: greedy hill-climb search and a genetic algorithm (GA).

Main Results:

• The genetically evolved Bayesian network achieved an Area Under the Curve (AUC) of 86.25%.
• This performance surpassed the greedy algorithm-developed network (65% AUC) and the naïve Bayesian classifier (84.75% AUC).
• The GA-based approach effectively identified dependencies among variables, unlike naïve Bayesian classifiers.

Conclusions:

• A genetically evolved Bayesian network is a superior method for age classification from ECG data.
• The methodology for evolving Bayesian classifiers can be broadly applied to discover network structures and variable dependencies.
• This approach holds significant promise for medical applications, including diagnosis and prediction.