Detection of subclinical rheumatic heart disease in children using a deep learning algorithm on digital stethoscope:...

Area of Science:

• Cardiology
• Artificial Intelligence
• Public Health

Background:

• Rheumatic heart diseases (RHDs) cause significant global morbidity and mortality.
• Timely secondary prophylaxis is crucial for reducing the RHD burden.

Purpose of the Study:

• Determine the frequency of subclinical RHD in school-going children in Karachi, Pakistan.
• Train a deep learning (DL) algorithm using digital auscultatory stethoscope (DAS) waveform data to predict subclinical RHD.

Main Methods:

• Recruit 1700 children (5-15 years) in Karachi, Pakistan.
• Collect sociodemographics, anthropometrics, medical history, and DAS data (phonocardiogram, electrocardiography).
• Perform handheld echocardiograms to identify mitral regurgitation (MR) or aortic regurgitation (AR); confirm with standard echocardiography.

Main Results:

• The DL algorithm will be trained using supervised learning with standard echocardiogram labels.
• The trained neural network will classify DAS data into definite RHD, borderline RHD, or normal categories.
• Statistical methods will confirm the significance of the findings.

Conclusions:

• This study will provide insights into the prevalence of subclinical RHD in a vulnerable population.
• The developed DL algorithm offers a potential tool for early RHD screening using a digital stethoscope.