An Intelligent Phonocardiography for Automated Screening of Pediatric Heart Diseases

Identifying pediatric heart murmurs and distinguishing innocent from pathologic using deep learning

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.

Identification of Congenital Valvular Murmurs in Young Patients Using Deep Learning-Based Attention Transformers and Phonocardiograms

One in every four newborns suffers from congenital heart disease (CHD) that causes defects in the heart structure. The current gold-standard assessment technique, echocardiography, causes delays in the diagnosis owing to the need for experts who vary markedly in their ability to detect and interpret pathological patterns. Moreover, echo is still causing cost difficulties for low- and middle-income countries. Here, we developed a deep learning-based attention transformer model to automate the detection of heart murmurs caused by CHD at an early stage of life using cost-effective and widely available phonocardiography (PCG). PCG recordings were obtained from 942 young patients at four major auscultation locations, including the aortic valve (AV), mitral valve (MV), pulmonary valve (PV), and tricuspid valve (TV), and they were annotated by experts as absent, present, or unknown murmurs. A transformation to wavelet features was performed to reduce the dimensionality before the deep learning stage for inferring the medical condition. The performance was validated through 10-fold cross-validation and yielded an average accuracy and sensitivity of 90.23 % and 72.41 %, respectively. The accuracy of discriminating between murmurs' absence and presence reached 76.10 % when evaluated on unseen data. The model had accuracies of 70 %, 88 %, and 86 % in predicting murmur presence in infants, children, and adolescents, respectively. The interpretation of the model revealed proper discrimination between the learned attributes, and AV channel was found important (score 0.75) for the murmur absence predictions while MV and TV were more important for murmur presence predictions. The findings potentiate deep learning as a powerful front-line tool for inferring CHD status in PCG recordings leveraging early detection of heart anomalies in young people. It is suggested as a tool that can be used independently from high-cost machinery or expert assessment.

Determining the incidence of heart malformations in neonates: A novel and clinically approved solution

Background

Screening for critical congenital heart defects should be performed as early as possible and is essential for saving the lives of children and reducing the incidence of undetected adult congenital heart diseases. Heart malformations remain unrecognized at birth in more than 50% of neonates at maternity hospitals. Accurate screening for congenital heart malformations is possible using a certified and internationally patented digital intelligent phonocardiography machine. This study aimed to assess the actual incidence of heart defects in neonates. A pre-evaluation of the incidence of unrecognized severe and critical congenital heart defects at birth in our well-baby nursery was also performed.

Methods

We conducted the Neonates Cardiac Monitoring Research Project (ethics approval number: IR-IUMS-FMD. REC.1398.098) at the Shahid Akbarabadi Maternity Hospital. This study was a retrospective analysis of congenital heart malformations observed after screening 840 neonates. Using a double-blind format, 840 neonates from the well-baby nursery were randomly chosen to undergo routine clinical examinations at birth and digital intelligent phonocardiogram examinations. A pediatric cardiologist performed echocardiography for each neonate classified as having abnormal heart sounds using an intelligent machine or during routine medical examinations. If the pediatric cardiologist requested a follow-up examination, then the neonate was considered to have a congenital heart malformation, and the cumulative incidence was calculated accordingly.

Results

The incidence of heart malformations in our well-baby nursery was 5%. Furthermore, 45% of heart malformations were unrecognized in neonates at birth, including one critical congenital heart defect. The intelligent machine interpreted innocent murmurs as healthy heart sound.

Conclusion

We accurately and cost-effectively screened for congenital heart malformations in all neonates in our hospital using a digital intelligent phonocardiogram. Using an intelligent machine, we successfully identified neonates with CCHD and congenital heart defects that could not be detected using standard medical examinations. The Pouya Heart machine can record and analyze sounds with a spectral power level lower than the minimum level of the human hearing threshold. Furthermore, by redesigning the study, the identification of previously unrecognized heart malformations could increase to 58%.

Deep learning-based computer-aided heart sound analysis in children with left-to-right shunt congenital heart disease

OBJECTIVE

The purpose of this study was to explore a new algorithm model capable of leverage deep learning to screen and diagnose specific types of left-to-right shunt congenital heart disease (CHD) in children.

METHODS

Using deep learning, screening models were constructed to identify 884 heart sound recordings from children with left-to-right shunt CHD. The most suitable model for each type was summarized and compared with expert auscultation. An exploratory analysis was conducted to assess whether there were correlations between heart sounds and left ventricular ejection fraction (LVEF), pulmonary artery pressure, and malformation size.

RESULTS

The residual convolution recurrent neural network (RCRnet) classification model had higher accuracy than other models with respect to atrial septal defect (ASD), ventricular septum defect (VSD), patent ductus arteriosus (PDA) and combined CHD, and the best auscultation sites were determined to be the 4th, 5th, 2nd and 3rd auscultation areas, respectively. The diagnostic results of this model were better than those derived from expert auscultation, with sensitivity values of 0.932-1.000, specificity values of 0.944-0.997, precision values of 0.888-0.997 and accuracy values of 0.940-0.994. Absolute Pearson correlation coefficient values between heart sounds of the four types of CHD and LVEF, right ventricular systolic pressure (RVSP) and malformation size were all less than 0.3.

CONCLUSIONS

The RCRnet model can preliminarily determine types of left-to-right shunt CHD and improve diagnostic efficiency, which may provide a new choice algorithmic CHD screening in children.

Diagnostic Accuracy of Machine Learning Models to Identify Congenital Heart Disease: A Meta-Analysis

Background: With the dearth of trained care providers to diagnose congenital heart disease (CHD) and a surge in machine learning (ML) models, this review aims to estimate the diagnostic accuracy of such models for detecting CHD. Methods: A comprehensive literature search in the PubMed, CINAHL, Wiley Cochrane Library, and Web of Science databases was performed. Studies that reported the diagnostic ability of ML for the detection of CHD compared to the reference standard were included. Risk of bias assessment was performed using Quality Assessment for Diagnostic Accuracy Studies-2 tool. The sensitivity and specificity results from the studies were used to generate the hierarchical Summary ROC (HSROC) curve. Results: We included 16 studies (1217 participants) that used ML algorithm to diagnose CHD. Neural networks were used in seven studies with overall sensitivity of 90.9% (95% CI 85.2-94.5%) and specificity was 92.7% (95% CI 86.4-96.2%). Other ML models included ensemble methods, deep learning and clustering techniques but did not have sufficient number of studies for a meta-analysis. Majority (n=11, 69%) of studies had a high risk of patient selection bias, unclear bias on index test (n=9, 56%) and flow and timing (n=12, 75%) while low risk of bias was reported for the reference standard (n=10, 62%). Conclusion: ML models such as neural networks have the potential to diagnose CHD accurately without the need for trained personnel. The heterogeneity of the diagnostic modalities used to train these models and the heterogeneity of the CHD diagnoses included between the studies is a major limitation.

A Framework for AI-Assisted Detection of Patent Ductus Arteriosus from Neonatal Phonocardiogram

The current diagnosis of Congenital Heart Disease (CHD) in neonates relies on echocardiography. Its limited availability requires alternative screening procedures to prioritise newborns awaiting ultrasound. The routine screening for CHD is performed using a multidimensional clinical examination including (but not limited to) auscultation and pulse oximetry. While auscultation might be subjective with some heart abnormalities not always audible it increases the ability to detect heart defects. This work aims at developing an objective clinical decision support tool based on machine learning (ML) to facilitate differentiation of sounds with signatures of Patent Ductus Arteriosus (PDA)/CHDs, in clinical settings. The heart sounds are pre-processed and segmented, followed by feature extraction. The features are fed into a boosted decision tree classifier to estimate the probability of PDA or CHDs. Several mechanisms to combine information from different auscultation points, as well as consecutive sound cycles, are presented. The system is evaluated on a large clinical dataset of heart sounds from 265 term and late-preterm newborns recorded within the first six days of life. The developed system reaches an area under the curve (AUC) of 78% at detecting CHD and 77% at detecting PDA. The obtained results for PDA detection compare favourably with the level of accuracy achieved by an experienced neonatologist when assessed on the same cohort.

Phonocardiogram Signal Processing for Automatic Diagnosis of Congenital Heart Disorders through Fusion of Temporal and Cepstral Features

Congenital heart disease (CHD) is a heart disorder associated with the devastating indications that result in increased mortality, increased morbidity, increased healthcare expenditure, and decreased quality of life. Ventricular Septal Defects (VSDs) and Arterial Septal Defects (ASDs) are the most common types of CHD. CHDs can be controlled before reaching a serious phase with an early diagnosis. The phonocardiogram (PCG) or heart sound auscultation is a simple and non-invasive technique that may reveal obvious variations of different CHDs. Diagnosis based on heart sounds is difficult and requires a high level of medical training and skills due to human hearing limitations and the non-stationary nature of PCGs. An automated computer-aided system may boost the diagnostic objectivity and consistency of PCG signals in the detection of CHDs. The objective of this research was to assess the effects of various pattern recognition modalities for the design of an automated system that effectively differentiates normal, ASD, and VSD categories using short term PCG time series. The proposed model in this study adopts three-stage processing: pre-processing, feature extraction, and classification. Empirical mode decomposition (EMD) was used to denoise the raw PCG signals acquired from subjects. One-dimensional local ternary patterns (1D-LTPs) and Mel-frequency cepstral coefficients (MFCCs) were extracted from the denoised PCG signal for precise representation of data from different classes. In the final stage, the fused feature vector of 1D-LTPs and MFCCs was fed to the support vector machine (SVM) classifier using 10-fold cross-validation. The PCG signals were acquired from the subjects admitted to local hospitals and classified by applying various experiments. The proposed methodology achieves a mean accuracy of 95.24% in classifying ASD, VSD, and normal subjects. The proposed model can be put into practice and serve as a second opinion for cardiologists by providing more objective and faster interpretations of PCG signals.

Fingerprinting of Doppler audio signals from the common carotid artery

Audio fingerprinting involves extraction of quantitative frequency descriptors that can be used for indexing, search and retrieval of audio signals in sound recognition software. We propose a similar approach with medical ultrasonographic Doppler audio signals. Power Doppler periodograms were generated from 84 ultrasonographic Doppler signals from the common carotid arteries in 22 dogs. Frequency features were extracted from each periodogram and included in a principal component analysis (PCA). From this 10 audio samples were pairwise classified as being either similar or dissimilar. These pairings were compared to a similar classification based on standard quantitative parameters used in medical ultrasound and to classification performed by a panel of listeners. The ranking of sound files according to degree of similarity differed between the frequency and conventional classification methods. The panel of listeners had an 88% agreement with the classification based on quantitative frequency features. These findings were significantly different from the score expected by chance (p < 0.001). The results indicate that the proposed frequency based classification has a perceptual relevance for human listeners and that the method is feasible. Audio fingerprinting of medical Doppler signals is potentially useful for indexing and search for similar and dissimilar audio samples in a dataset.

Digital stethoscopes in paediatric medicine

AIM

To explore, synthesise and discuss currently available digital stethoscopes (DS) and the evidence for their use in paediatric medicine.

METHODS

Systematic review and narrative synthesis of digital stethoscope use in paediatrics following searches of OVID Medline, Embase, Scopus, PubMed and Google Scholar databases.

RESULTS

Six digital stethoscope makes were identified to have been used in paediatric focused studies so far. A total of 25 studies of DS use in paediatrics were included. We discuss the use of digital stethoscope technology in current paediatric medicine, comment on the technical properties of the available devices, the effectiveness and limitations of this technology, and potential uses in the fields of paediatrics and neonatology, from telemedicine to computer-aided diagnostics.

CONCLUSION

Further validation and testing of available DS devices is required. Comparison studies between different types of DS would be useful in identifying strengths and flaws of each DS as well as identifying clinical situations for which each may be most appropriately suited.

The first 200 years of cardiac auscultation and future perspectives

Cardiac auscultation - even with its limitations - is still a valid and economical technique for the diagnosis of cardiovascular diseases, and despite the growing demand for sophisticated imaging techniques, clinical use of the stethoscope in medical practice has not yet been abandoned. In 1816, René-Théophile-Hyacinthe Laënnec invented the stethoscope, while examining a young woman with suspected heart disease, giving rise to mediated auscultation. He described in detail several heart and lung sounds, correlating them with postmortem pathology. Even today, a correct interpretation of heart sounds, integrated with the clinical history and physical examination, allows to detect properly most of the structural heart abnormalities or to evaluate them in a differential diagnosis. However, the lack of organic teaching of auscultation and its inadequate practice have a negative impact on the clinical competence of physicians in training, also reflecting a diminished academic interest in physical semiotic. Medical simulation could be an effective instructional tool in teaching and deepening auscultation. Handheld ultrasound devices could be used for screening or for integrating and improving auscultatory abilities of physicians; the electronic stethoscope, with its new digital capabilities, will help to achieve a correct diagnosis. The availability of innovative representations of the sounds with phono- and spectrograms provides an important aid in diagnosis, in teaching practice and pedagogy. Technological innovations, despite their undoubted value, must complement and not supplant a complete physical examination; clinical auscultation remains an important and cost-effective screening method for the physicians in cardiorespiratory diagnosis. Cardiac auscultation has a future, and the stethoscope has not yet become a medical heirloom.

A Deep Machine Learning Method for Classifying Cyclic Time Series of Biological Signals Using Time-Growing Neural Network

This paper presents a novel method for learning the cyclic contents of stochastic time series: the deep time-growing neural network (DTGNN). The DTGNN combines supervised and unsupervised methods in different levels of learning for an enhanced performance. It is employed by a multiscale learning structure to classify cyclic time series (CTS), in which the dynamic contents of the time series are preserved in an efficient manner. This paper suggests a systematic procedure for finding the design parameter of the classification method for a one-versus-multiple class application. A novel validation method is also suggested for evaluating the structural risk, both in a quantitative and a qualitative manner. The effect of the DTGNN on the performance of the classifier is statistically validated through the repeated random subsampling using different sets of CTS, from different medical applications. The validation involves four medical databases, comprised of 108 recordings of the electroencephalogram signal, 90 recordings of the electromyogram signal, 130 recordings of the heart sound signal, and 50 recordings of the respiratory sound signal. Results of the statistical validations show that the DTGNN significantly improves the performance of the classification and also exhibits an optimal structural risk.

A Visualization System for Interactive Exploration of the Cardiac Anatomy

Because of the complex and fine structure, visualization of the heart still remains a challenging task, which makes it an active research topic. In this paper, we present a visualization system for medical data, which takes advantage of the recent graphics processing unit (GPU) and can provide real-time cardiac visualization. This work focuses on investigating the anatomical structure visualization of the human heart, which is fundamental to the cardiac visualization, medical training and diagnosis assistance. Several state-of-the-art cardiac visualization methods are integrated into the proposed system and a task specified visualization method is proposed. In addition, auxiliary tools are provided to generate user specified visualization results. The contributions of our work lie in two-fold: for doctors and medical staff, the system can provide task specified visualization with interactive visualization tools; for researchers, the proposed platform can serve as a baseline for comparing different rendering methods and can easily incorporate new rendering methods. Experimental results show that the proposed system can provide favorable cardiac visualization results in terms of both effectiveness and efficiency.