Deep Learning Algorithm for Heart Valve Diseases Assisted Diagnosis

Heart sounds are mainly the expressions of opening and closing heart valves. Some produced by interruption laminar blood flow as it turns into turbulent flow, which is explained abnormal functioning The analysis phonocardiographic signals has made possible to indicate that normal pathological records differ from each other concerning both temporal spectral features. present work describes design implementation based on deep neural networks learning for binary multiclass classification four common valvular pathologies sounds. For feature extraction, three different techniques were considered: Discrete Wavelet Transform, Continuous Transform Mel Frequency Cepstral Coefficients. performance approaches reached F1 scores higher than 98% specificities in “Normal” class up 99%, considers cases can be misclassified normal. These results place a highly competitive proposal generation systems assisted diagnosis.