Bayesian classification applied to strain in arrythmogenic left-ventricle cardiomyopathy

摘要

Arrhythmogenic cardiomyopathy (AC) is a rare disease associated with ventricular arrhythmias and sudden cardiac death. While AC of the right ventricle has been more extensively studied, exclusive left-ventricle involvement needs to be better characterized. Myocardial strain, obtained by feature tracking, provide insight into its biomechanical behavior. To characterize it, multivariate classifiers can be applied. The sample consisted of 13 AC-LV and 13 non-carriers of the mutation. The feature tracking algorithm of Circle cvi^{42was applied to the cardiac magnetic resonance of each patient. A Naïve Bayes classifier with a feature subset selection method was applied to the parameters of peak strain, strain rate, displacement and velocity. We obtained an accuracy of 90% in NB and we arrived to 93% for CFS-NB. The strain parameters selected by the FSS algorithm were three: longitudinal peak strain and peak systolic and diastolic velocities. In all the selected features, AC-LV patients had smaller values as controls. In conclusion, myocardial strain is affected in AC-LV patients. Naïve Bayes classifiers allow obtaining a good discriminating accuracy among groups.}