Optimal Kernels application to improve late ventricular activity detection

Resumen

Sometimes when a myocardium infarct occurs, small abnormalities in conduction are present over the infarcted zone. These components are known as Ventricular Late Potentials (VLP) and are associated with ventricular arrhythmias and sudden cardiac death. They are components of ventricular conduction activity that are attenuated, fragmented and delayed over the QRS complex of an electrocardiogram (ECG). VLP is often used as non-invasive markers of arrhythmia risk and while their detection is difficult, there are non-invasive methods proposed for improved detection. The classical time domain method is the most often used for VLP detection in the analysis of high resolution ECG (HRECG) on post-infarction patients. Nonetheless, it brings low predictive values, high sensibility to noise and excludes in its analysis patients with bundle branch blockage. In this paper, the different morphologies of VLP are used for deducting a bi-dimensional Kernel in the time-frequency domain, so that it can be adapted to changing VLP structures according to each post-infarct patient. Also, both the reduction of false negatives and an increase in true positives of the automatic diagnosis can be achieved. A database of 132 HRECG signals was analyzed and a substantial increase in predictive values was obtained over diagnostics. In the analysis, attenuated sensibility to noise compared to the classical temporal domain method was also shown.