Implementation of a fuzzy prototype-based machine learning method to predict myocardial infarction from coronary angiography.

ABSTRACT

Formal knowledge on the predictive value of morphological angiographic factors is lacking to estimate the risk of myocardial infarction. This article presents a computer system for predicting the incidence of myocardial infarction from angiographic morphological descriptions of coronary lesions. The system includes two phases. The learning phase consists in extracting from a large database of described stenoses two classes represented by one or several fuzzy prototypes. One class corresponds to stenoses leading to infarction and the other to stenoses not leading to that event. The evaluation phase consists in classifying a stenosis according to its morphological characteristics in one of these two classes. The learning method is based on a fuzzy supervised Machine Learning algorithm that combines some aspects of the K-nearest neighbours clustering approach with a defined measure of similarity, and a prototype induction function from the most similar stenoses, taking into account their degree of typicality. The current results of the evaluation phase to correctly predicted X% stenoses for their risk of myocardial infarction. This article emphasizes the feasibility of the approach, however, the learning phase relies on some heuristics that should be validated to get a formal evaluation of the system.