Citation: Flores-Alonso, S.I.;
Tovar-Corona, B.; Luna-García, R.
Deep Learning Algorithm for Heart
Valve Diseases Assisted Diagnosis.
Appl. Sci. 2022, 12, 3780. https://
doi.org/10.3390/app12083780
Academic Editors: Keun Ho Ryu and
Nipon Theera-Umpon
Received: 25 February 2022
Accepted: 6 April 2022
Published: 8 April 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Article
Deep Learning Algorithm for Heart Valve Diseases
Assisted Diagnosis
Santiago Isaac Flores-Alonso
1,†
, Blanca Tovar-Corona
2,†
and René Luna-García
1,
*
,†
1
CIC-IPN, Ciudad de México 07738, Mexico; sfloresa2010@alumno.ipn.mx
2
UPIITA-IPN, Ciudad de México 07340, Mexico; bltovar@ipn.mx
* Correspondence: rlunag@ipn.mx
† These authors contributed equally to this work.
Abstract:
Heart sounds are mainly the expressions of the opening and closing of the heart valves.
Some sounds are produced by the interruption of laminar blood flow as it turns into turbulent flow,
which is explained by abnormal functioning of the valves. The analysis of the phonocardiographic
signals has made it possible to indicate that the normal and pathological records differ from each
other concerning both temporal and spectral features. The present work describes the design and
implementation based on deep neural networks and deep learning for the binary and multiclass clas-
sification of four common valvular pathologies and normal heart sounds. For feature extraction, three
different techniques were considered: Discrete Wavelet Transform, Continuous Wavelet Transform
and Mel Frequency Cepstral Coefficients. The performance of both approaches reached F1 scores
higher than 98% and specificities in the “Normal” class of up to 99%, which considers the cases that
can be misclassified as normal. These results place the present work as a highly competitive proposal
for the generation of systems for assisted diagnosis.
Keywords:
deep learning; CWT; deep neural networks; DWT; MFCCs; phonocardiography; valvular
disease
1. Introduction
Cardiovascular diseases occupy first place among the causes of death around the
world, according to the World Health Organization (WHO) [
1
]. Heart valve diseases (HVD)
are also found in these figures, where moderate or severe valvular abnormalities are notably
common in the adult population and increase their presence as the individuals age [2].
To examine the condition of the heart valves, the most common medical practice is
auscultation, which consists of listening to acoustic characteristics directly via the patient’s
chest wall using a stethoscope. These heart sounds could be interpreted as the acoustic
expression of the opening and closing of the four heart valves—tricuspid, mitral, pulmonary
and aortic—where the muscular contraction that drives the blood from one cavity to
another generates a high acceleration and delay of the blood flow, causing a pressure
difference [
3
,
4
]. Its normal physiological functioning is always unidirectional, which allows
the correct circulation of blood through the cardiovascular circuit. However, some sounds
are produced by the interruption of laminar blood flow by turning into turbulent flow,
which is explained by abnormal and pathological functioning of the heart valves.
The cardiac cycle is composed of two phases: the systole, during which the ventricles
contract and drive blood to the blood vessels, and the diastole, in which the ventricles are
filled.The systole begins with the closure of the mitral and triscuspid valves, producing the
first heart sound or S1, while the diastole starts with the closure of the aortic and pulmonic
valves, producing the second heart sound or S2. In addition to S1 and S2, extra sounds
could be present during the cardiac cycle, which may indicate an abnormality [
5
,
6
]. The
duration of the noise varies depending on the valvular abnormality [
7
]. Figure 1 illustrates
Appl. Sci. 2022, 12, 3780. https://doi.org/10.3390/app12083780 https://www.mdpi.com/journal/applsci
