Método no invasivo de detección de fibrilación ventricular mediante análisis tiempo-frecuencia

Abstract

Sudden cardiac death is one of the main causes of death in Europe and North America, since it presents itself in numerous cardiopathies, especially ischemic and the so-called miocardiopathies. Ventricular fibrillation (VF) is the most common cause since it leads to a cardiorespiratory failure as the heart fails to act mechanically. Early detection is key to avoid sudden death or irreversible damage. The current work proposes a detection technique based on the time-frequency domain and oriented to the real-time detection of the onset of ventricular fibrillation. The technique could be incorporated into cardiac monitors and external defibrillators to use in intensive care hospital units, ambulatory centers, public places etc. It provides not only help for the cardiologist to make a diagnosis, but also for nonspecialized sanitary personnel in circumstances where a specialist is not present. Signal acquisition is non-invasive using a single cardiac derivation and surface electrodes. In addition, to increase the efficiency of the system, we propose the use of time-domain parameters that reduce the computational load. The registers used to correspond to half-hour measurements extracted from standard databases. The time-frequency distribution used is the Pseudo Wigner-Ville distribution. We propose different detection algorithms, we first select the most representative parameters and then apply them to detection systems based on discriminant analysis, detection trees and neural networks. We work mainly on discriminating VF and certain taquicardies that produce the most usual detection errors. The results obtained offer a specificity of between 75% and 99% and a sensibility of between 83% and 96%.