Congenital heart disease classification using phonocardiograms: a scalable screening tool for diverse environments

Q: What is the clinical evidence from this study?

Study design: Cross-Sectional. Population: Congenital heart disease (CHD) (n=751). Intervention: Phonocardiogram (PCG) deep learning classification model vs. Clinical reference standard (echocardiography, chest X-ray, 12-lead ECG, expert assessment). Primary outcome: Detection of congenital heart disease (Diagnostic Accuracy).

Q: What does this research mean for the field?

A deep learning model analyzing phonocardiogram signals detects pediatric congenital heart disease with high diagnostic accuracy, maintaining robust performance even on low-quality recordings and single auscultation sites. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Abdul Jabbar; Ethan Grooby; Jack Crozier; Alexander Gallon; Vivian Pham; Khawza I Ahmad; Raqibul Mostafa; Ahsan Habib Khandoker; Faezeh Marzbanrad

doi:10.1038/s41598-026-52029-1

September 17, 2026Scientific ReportsOpen Access

Congenital heart disease classification using phonocardiograms: a scalable screening tool for diverse environments

View Full Paper

Ask AI

Bookmark

Share

Key result

Deep learning analysis of phonocardiograms detects congenital heart disease with ~94% accuracy.

n=751

Why the study?

Early detection of congenital heart disease in infancy and childhood is critical, and scalable screening tools suitable for diverse and resource-limited environments are needed.

Does a deep learning model analyzing raw phonocardiogram signals improve automated detection of congenital heart disease in pediatric populations in resource-limited settings?

Key Points

To develop and validate an automated deep learning framework for detecting congenital heart disease from raw phonocardiogram audio signals across diverse, resource-limited clinical environments.
Engineered a deep learning architecture using transfer learning from murmur detection to classify congenital heart disease directly from raw phonocardiogram recordings without manual acoustic segmentation.
Trained and evaluated the model on a primary pediatric dataset from Bangladesh and tested generalization across external datasets, including PhysioNet Challenge 2022 and PhysioNet Challenge 2016.
Assessed performance across single versus four combined auscultation chest sites (aortic, mitral, pulmonary, and tricuspid valves) and on recordings classified as low quality or non-diagnostic by cardiologists.
On the main Bangladesh dataset, the model achieved an accuracy of 94.1%, sensitivity of 92.7%, specificity of 96.3%, and AUROC of 98.1%.
On low-quality recordings deemed non-diagnostic by cardiologists, the algorithm demonstrated an accuracy of 80.5%, sensitivity of 79.7%, specificity of 81.5%, and AUROC of 87.7%.
Testing across single auscultation sites achieved an accuracy of 85.68% and an AUROC of 91.6% at the aortic valve site, while external testing on the unseen PhysioNet 2022 split yielded 91.6% accuracy, 92.2% sensitivity, and 88.9% specificity.

Study Design

Type

Cross-Sectional (n=751)

Multicenter

Yes

Structured PICO

Does a deep learning model analyzing raw phonocardiogram signals improve automated detection of congenital heart disease in pediatric populations in resource-limited settings?

P

Population

751 predominantly pediatric and adolescent patients (456 with congenital heart disease and 295 healthy controls) from Bangladesh were evaluated to develop and test a deep learning model for CHD detection.

E

Exposure

A deep learning model (InceptionTime-based convolutional neural network with residual connections and transfer learning) classifying congenital heart disease directly from raw phonocardiogram (PCG) signals without requiring segmentation, collected via an Eko DUO ECG + Digital Stethoscope.

C

Comparator

Alternative machine learning models (classical SVM with handcrafted features, ResNet1D, and a 1D CNN trained from scratch) as well as top-performing models from previous PhysioNet Challenges.

O

Outcome

Diagnostic performance for classifying congenital heart disease, measured by accuracy, sensitivity, specificity, and Area Under the Receiver Operating Characteristic (AUROC) curve.surrogate

A scalable deep learning model operating directly on raw phonocardiogram signals can accurately detect congenital heart disease, providing a cost-effective digital screening tool for resource-limited environments.

Limitations

Model underperformed in terms of sensitivity when tested on adult populations (PhysioNet 2016 dataset).
Decreased performance on low-quality/noisy recordings, highlighting the need for noise removal and signal quality assessment algorithms.
Analysis was limited to 15-second PCG signal segments rather than full-length recordings.
The model only performed binary classification (CHD vs non-CHD) rather than classifying specific CHD types.
Lower sensitivity (55.5%) when tested on the adult-predominant PhysioNet 2016 dataset, likely because the model was trained on pediatric data
Labels in external datasets may refer to other types of heart diseases rather than explicitly congenital heart disease

Implication

May support AI phonocardiogram screening for congenital heart disease; leaves open prospective validation across diverse settings.

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 transferability 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.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Jabbar et al. (2026) conducted a cross-sectional in Congenital heart disease (CHD) (n=751). Phonocardiogram (PCG) deep learning classification model vs. Clinical reference standard (echocardiography, chest X-ray, 12-lead ECG, expert assessment) was evaluated on Detection of congenital heart disease (Diagnostic Accuracy). A deep learning model analyzing phonocardiogram signals achieved 94.1% accuracy, 92.7% sensitivity, and 96.3% specificity in detecting congenital heart disease in a primary pediatric dataset.

synapsesocial.com/papers/6ad0fd3315427f6a840405ee https://doi.org/10.1038/s41598-026-52029-1

View Full Paper

Ask AI

Bookmark

Share

Congenital heart disease classification using phonocardiograms: a scalable screening tool for diverse environments

Key Points

Study Design

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

Congenital heart disease classification using phonocardiograms: a scalable screening tool for diverse environments

Key Points

Study Design

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading