HeartSound

Description

Android mobile app for heart sound audio classification. It classifies heart sound into two groups: abnormal and normal sound. HeartSound is standalone machine learning app that uses Physionet Challenge Dataset 2016 for model training. The app can achieve 82.77% in accuracy.

Background

Screening abnormal heart sound by lay persons may help detecting heart diseases earlier. Machine learning can be used to develop mobile device for helping them doing this screening easily. Fortunately, there is The Physionet Challenge Dataset 2016 that contains of 3,240 heart sound files consisting of 665 abnormal sound files and 2,575 normal sound files. With Tensorflow technology and this data set makes it possible to develop machine learning model for heart sound audio classification.

Features

This app can classify heart sound in two mode:

  1. Real-time audio classification.

  2. Audio file classification.

Result statistics

HeartSound app is tested with test group data. The statistics are as follow:

Test group characteristics:

  • Total 325 cases

  • Prevalence of abnormal sound: 20.62%

Key statistics:

  • p-value: 0.000

  • Sensitivity: 71.64%

  • Specificity: 85.66%

  • Precision: 56.47%

  • Negative Predictive Value: 92.08%

  • Accuracy: 82.77%

Heart sound file specification

WAV mono

Screenshots

Opening stage

Real Time Classification

File based Classification

Other software and equipment may be needed

1. To get good quality of heart sound, you may need electronic stethoscope that can be connected to your android device. It can be commercial product or you can make it by yourself. The following picture is DIY electronic stethoscope that connects condenser microphone with the head of stethoscope.

DIY electronic stethoscope

2. To record heart sound, you may need audio recorder app installed on your device. This can help you recording and converting heart sound into file. To use with HeartSound app, the file created must be in WAV format with MONO channel and it will be better to use 16 kHz of sampling rate.

Where to listen for Heart Sounds by Medzcool

History

  • version 1.0.1 released ( 19 August 2021)

  • version 1.0 released