Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Congenital Heart Disease Classification Using Phonocardiograms: A Scalable Screening Tool for Diverse Environments

Abdul Jabbar, Ethan Grooby, Jack Crozier, Alexander Gallon, Vivian Pham, Khawza I Ahmad, Md Hassanuzzaman, Raqibul Mostafa, Ahsan H. Khandoker, Faezeh Marzbanrad

TL;DR

This work tackles the urgent need for scalable CHD screening in low-resource settings by leveraging PCG signals collected via a digital stethoscope and a deep learning model. The authors pre-train a 1D CNN with Inception-Time-like architecture on the PhysioNet 2022 murmurs task and fine-tune it on a large Bangladesh CHD dataset, achieving 94.1% accuracy and 98.1% AUROC on the main data, with solid transferability to public datasets (AUROC up to 94.4%). They analyze performance across auscultation sites, showing single-site viability (AV site ~85.7% accuracy) and robust results even for low-quality recordings (≈80% accuracy), underscoring practical utility in LMICs. The study demonstrates that a scalable AI-driven digital stethoscope-based screening tool can augment clinical decision support, enabling early CHD detection where echocardiography access is limited, while highlighting limitations and directions for multi-class classification and enhanced noise handling.

Abstract

Congenital heart disease (CHD) is a critical condition that demands early detection, particularly in infancy and childhood. This study presents a deep learning model designed to detect CHD using phonocardiogram (PCG) signals, with a focus on its application in global health. We evaluated our model on several datasets, including the primary dataset from Bangladesh, achieving a high accuracy of 94.1%, sensitivity of 92.7%, specificity of 96.3%. The model also demonstrated robust performance on the public PhysioNet Challenge 2022 and 2016 datasets, underscoring its generalizability to diverse populations and data sources. We assessed the performance of the algorithm for single and multiple auscultation sites on the chest, demonstrating that the model maintains over 85% accuracy even when using a single location. Furthermore, our algorithm was able to achieve an accuracy of 80% on low-quality recordings, which cardiologists deemed non-diagnostic. This research suggests that an AI- driven digital stethoscope could serve as a cost-effective screening tool for CHD in resource-limited settings, enhancing clinical decision support and ultimately improving patient outcomes.

Congenital Heart Disease Classification Using Phonocardiograms: A Scalable Screening Tool for Diverse Environments

TL;DR

This work tackles the urgent need for scalable CHD screening in low-resource settings by leveraging PCG signals collected via a digital stethoscope and a deep learning model. The authors pre-train a 1D CNN with Inception-Time-like architecture on the PhysioNet 2022 murmurs task and fine-tune it on a large Bangladesh CHD dataset, achieving 94.1% accuracy and 98.1% AUROC on the main data, with solid transferability to public datasets (AUROC up to 94.4%). They analyze performance across auscultation sites, showing single-site viability (AV site ~85.7% accuracy) and robust results even for low-quality recordings (≈80% accuracy), underscoring practical utility in LMICs. The study demonstrates that a scalable AI-driven digital stethoscope-based screening tool can augment clinical decision support, enabling early CHD detection where echocardiography access is limited, while highlighting limitations and directions for multi-class classification and enhanced noise handling.

Abstract

Congenital heart disease (CHD) is a critical condition that demands early detection, particularly in infancy and childhood. This study presents a deep learning model designed to detect CHD using phonocardiogram (PCG) signals, with a focus on its application in global health. We evaluated our model on several datasets, including the primary dataset from Bangladesh, achieving a high accuracy of 94.1%, sensitivity of 92.7%, specificity of 96.3%. The model also demonstrated robust performance on the public PhysioNet Challenge 2022 and 2016 datasets, underscoring its generalizability to diverse populations and data sources. We assessed the performance of the algorithm for single and multiple auscultation sites on the chest, demonstrating that the model maintains over 85% accuracy even when using a single location. Furthermore, our algorithm was able to achieve an accuracy of 80% on low-quality recordings, which cardiologists deemed non-diagnostic. This research suggests that an AI- driven digital stethoscope could serve as a cost-effective screening tool for CHD in resource-limited settings, enhancing clinical decision support and ultimately improving patient outcomes.

Paper Structure

This paper contains 29 sections, 1 equation, 4 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Bangladesh Dataset
  4. Dataset and Patient Enrolment
  5. Data Preparation and Labelling
  6. PhysioNet Challenge 2022 Dataset
  7. Dataset and Patient Enrolment
  8. Data Preparation and Labelling
  9. PhysioNet Challenge 2016 Dataset
  10. Preprocessing
  11. Down-sampling
  12. Normalization
  13. Base Model Architecture
  14. Model Initialization
  15. Fine-tuning for CHD Detection
  16. ...and 14 more sections

Figures (4)

  • Figure 1: Block diagram of proposed model for CHD detection
  • Figure 2: Performance Evaluation on the Unseen Patient-Wise Split of the main Dataset: (a) Receiver Operating Characteristic (ROC) Curve (b) Precision-Recall Curve
  • Figure 3: Performance Evaluation on the Unseen Patient-Wise Split of the PhysioNet challenge 2022 dataset: (a) Receiver Operating Characteristic (ROC) Curve (b) Precision-Recall Curve
  • Figure 4: Model performance evaluation based on training with each individual PCG channel compared to the original combined PCG model. The channels include the aortic valve (AV), mitral valve (MV), pulmonary valve (PV), and tricuspid valve (TV). The performance of the combined model is represented by the final bar in each plot