Related Experiment Video
Updated: Jun 26, 2025
A Modified Sonographic Algorithm for Image Acquisition in Life-Threatening Emergencies in the Critically Ill Newborn
Published on: April 7, 2023
Identifying pediatric heart murmurs and distinguishing innocent from pathologic using deep learning
George Zhou1, Candace Chien2, Justin Chen3
1Weill Cornell Medicine, New York, NY 10021, USA.
Insights
This study introduces novel deep learning algorithms for classifying pediatric heart sounds, achieving high accuracy in distinguishing normal sounds from innocent and pathologic murmurs. These advanced methods outperform current standards, offering improved diagnostic potential.
Area of Science:
- Artificial Intelligence
- Cardiology
- Medical Imaging
Background:
- Accurate classification of pediatric heart sounds is crucial for timely diagnosis and treatment.
- Current methods for heart sound analysis often lack the precision needed for complex pediatric cases.
- Deep learning offers potential for enhanced accuracy in analyzing subtle variations in heart sounds.
Purpose of the Study:
- To develop and evaluate deep learning algorithms for multi-class classification of pediatric heart sounds.
- To differentiate between normal heart sounds, innocent murmurs, and pathologic murmurs using advanced AI techniques.
- To compare the performance of novel Vision Transformer models against traditional Convolutional Neural Network (CNN) approaches.
Main Methods:
- Utilized a dataset of pediatric heart sounds, including normal, innocent, and pathologic murmurs, augmented with public data.
- Developed two novel approaches using a Vision Transformer trained on Gramian Angular Field (GAF) and Markov Transition Field (MTF) image representations.
- Benchmarked Vision Transformer models against a ResNet-50 CNN trained on spectrogram images.
Main Results:
- The Vision Transformer models consistently outperformed the ResNet-50 CNN across all tested image representations.
- The Gramian Angular Field (GAF) representation demonstrated superior performance for pediatric heart sound classification.
- The best model achieved high Area Under the Curve (AUC) values: 0.92 for normal sounds, 0.83 for innocent murmurs, and 0.88 for pathologic murmurs.
Conclusions:
- Novel deep learning methods, particularly Vision Transformers with GAF, significantly improve pediatric heart sound classification accuracy.
- This study presents the first demonstration of multi-class classification for pediatric murmurs using deep learning.
- The developed models offer a more explainable and interpretable approach, potentially increasing clinician trust and adoption in clinical practice.
Objective:
To develop a deep learning algorithm to perform multi-class classification of normal pediatric heart sounds, innocent murmurs, and pathologic murmurs.
Methods:
We prospectively enrolled children under age 18 being evaluated by the Division of Pediatric Cardiology. Parents provided consent for a deidentified recording of their child's heart sounds with a digital stethoscope. Innocent murmurs were validated by a pediatric cardiologist and pathologic murmurs were validated by echocardiogram. To augment our collection of normal heart sounds, we utilized a public database of pediatric heart sound recordings (Oliveira, 2022). We propose two novel approaches for this audio classification task. We train a vision transformer on either Markov transition field or Gramian angular field image representations of the frequency spectrum. We benchmark our results against a ResNet-50 CNN trained on spectrogram images.
Results:
Our final dataset consisted of 366 normal heart sounds, 175 innocent murmurs, and 216 pathologic murmurs. Innocent murmurs collected include Still's murmur, venous hum, and flow murmurs. Pathologic murmurs included ventricular septal defect, tetralogy of Fallot, aortic regurgitation, aortic stenosis, pulmonary stenosis, mitral regurgitation and stenosis, and tricuspid regurgitation. We find that the Vision Transformer consistently outperforms the ResNet-50 on all three image representations, and that the Gramian angular field is the superior image representation for pediatric heart sounds. We calculated a one-vs-rest multi-class ROC curve for each of the three classes. Our best model achieves an area under the curve (AUC) value of 0.92 ± 0.05, 0.83 ± 0.04, and 0.88 ± 0.04 for identifying normal heart sounds, innocent murmurs, and pathologic murmurs, respectively.
Conclusion:
We present two novel methods for pediatric heart sound classification, which outperforms the current standard of using a convolutional neural network trained on spectrogram images. To our knowledge, we are the first to demonstrate multi-class classification of pediatric murmurs. Multiclass output affords a more explainable and interpretable model, which can facilitate further model improvement in the downstream model development cycle and enhance clinician trust and therefore adoption.
Related Concept Videos
01:15Heart Sounds
Auscultation is the process of listening to these internal body sounds using a stethoscope. The heart produces four types of sounds, but only two—S1 and S2—can usually be heard with a stethoscope.
S1, also known as the "lub" sound, is caused by the closure of atrioventricular (A-V)...
01:25Cardiovascular System Abnormal Findings II: Auscultation
Abnormal Heart Sounds
Gallops:
01:13Special considerations while measuring pulse
01:20Imaging Studies for Cardiovascular System II:Types of Echocardiography
Types of Echocardiography
Transthoracic Echocardiography (TTE)
TTE is the most common type of echocardiogram which involves placing a transducer on the patient's chest, emitting sound waves to create heart images. TTE is invaluable for evaluating the heart's size, structure, and motion, making it particularly useful for...