Phonocardiographic Classification of Mechanical Heart Valves Using Artificial Neural Networks

Abstract

Nowadays mechanical heart valves (MHVs) maintain a primary role for the surgical treatment of valvulopaties. MHVs are suitable for those patients who can be treated with anticoagulant therapy and whose life expectancy is longer than 10-15 years, that is the estimated durability of a valvular bioprosthesis. Though, the implanted mechanical valves and the anticoagulation level have to be regularly monitored to avoid thromboembolitic complications. This study presents an innovative approach for the early detection of MHVs dysfunctions. Closure sounds of 5 different bileaflet valves, both normofunctioning and thrombotic, were recorded during in vitro simulations under different working conditions using the Sheffield Pulse Duplicator; their power spectra were then used to train artificial neural networks of specific topology. The resulting high classification performance of the networks and the ongoing in vivo application to St. Jude Regent valves, confirm the possibility to use these classifiers, after an appropriate clinical validation, to identify bileaflet valves requiring further medical examinations.