Diagnosing mitral valve prolapse by improving the predictive power of classifiers

Area of Science:

• Cardiology
• Medical Informatics
• Computational Intelligence

Background:

• Mitral valve prolapse (MVP) is a prevalent cardiac valvular abnormality in industrialized nations, associated with risks including sudden death.
• Accurate diagnosis of MVP is crucial for patient management and risk stratification.
• Existing diagnostic methods may benefit from improved feature selection for enhanced predictive accuracy.

Purpose of the Study:

• To evaluate various feature selection mechanisms for improving the diagnostic accuracy of classifiers for mitral valve prolapse (MVP).
• To investigate the efficacy of classical greedy approaches, a genetic algorithm (GA), and a cellular automaton (CA) combined with GA (CA-GA) for feature selection in MVP diagnosis.
• To assess the ability of feature selection methods to generalize knowledge for MVP diagnosis.

Main Methods:

• Employed classical greedy feature selection methods: forward selection and backward elimination.
• Utilized a genetic algorithm (GA) for feature selection.
• Implemented a novel cellular automaton (CA) integrated with GA (CA-GA) for data transformation and feature selection within the knowledge discovery process.

Main Results:

• The CA-GA approach demonstrated superior performance compared to classical greedy feature selection methods.
• Feature subsets generated by GA and CA-GA approaches were highly effective when used with a decision tree classifier for MVP diagnosis, achieving the highest overall class accuracy.
• The CA and GA methods successfully generalized significant knowledge pertinent to MVP diagnosis.

Conclusions:

• The CA-GA approach represents a powerful tool for feature selection in the context of medical diagnosis, specifically for mitral valve prolapse.
• Integrating CA with GA enhances the predictive capabilities of classifiers, leading to more accurate MVP diagnosis.
• These advanced computational methods offer valuable insights and improve the generalization of knowledge for complex medical conditions like MVP.