Data Processing

Libraries of heart and lung sounds each are sourced online. The heart sound library is taken from the PhysioNet Challenge 2016, while the lung sound library is taken from ICBHI (International Conference on Biomedical and Health Informatics) Challenge 2017.

Using MATLAB, each of the sound libraries are machine trained to distinguish features within heart and lung sounds, to determine the normality of the heart and lung sounds. This is done through feature extraction and selection, as well as Bayesian optimisation, latter of which is optimisation that does not assume any functional forms. Misclassification costs are also considered in Bayesian optimisation for training. After model training, the algorithms are integrated into an app generated by MATLAB App Designer. The app is compiled into a standalone .exe application, which processes lung and heart sound data provided by the user.

Below outlines the flowchart on the creation of the classification model:

In the compiled .exe application, the application is divided into 2 tabs, first tab which processes heart sounds and another which processes lung sounds. Each tab has only 1 clickable button which is to upload a file. A wav file will have to be selected for data processing. After successful data processing, the spectrogram for the .wav file will be displayed and the diagnosis of the sound will be displayed, as illustrated below: