ENACT-Heart – ENsemble-based Assessment Using CNN and Transformer on Heart Sounds

Jiho Han 1, Adnan Shaout · 2025

Read ENACT-Heart – ENsemble-based Assessment Using CNN and Transformer on Heart Sounds by Jiho Han 1, Adnan Shaout (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-16914.txt.

Introduction

Jiho Han and Adnan Shaout blend, in 2025 work, two net kinds to sort heart sounds. One net reads local shapes; the other, born for images, reads the whole at once. A mix-of-experts frame lets each do its best work, then pools the calls.

The blend hit near ninety-eight percent true calls. That beat each net on its own, the shape net by two points and the whole-read net by near four. The pair frame the win as proof that pooled experts lift sound-based heart checks, a cheap and wide tool for heart watch.

Catalog record

CREATORJiho Han 1, Adnan Shaout
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-16914

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

ENACT-Heart – ENsemble-based Assessment Using CNN and Transformer on Heart Sounds Jiho Han 1, Adnan Shaout Abstract This study explores the application of Vision Transformer (ViT) principles in audio analysis, specifically focusing on heart sounds. This paper introduces ENACT-Heart – a novel ensemble approach that leverages the complementary strengths of Convolutional Neural Networks (CNN) and ViT through a Mixture of Experts (MoE) framework, achieving a remarkable classification accuracy of 97.52%. This outperforms the individual contributions of ViT (93.88%) and CNN (95.45%), demonstrating the potential for enhanced diagnostic accuracy in cardiovascular health monitoring. These results demonstrate the potential of ensemble methods in enhancing classification performance for cardiovascular health monitoring and diagnosis. I Introduction Cardiac diagnostics have undergone remarkable advancements over the centuries, yet the analysis of heart sounds remains a fundamental aspect of assessing cardiovascular health. These sounds, primarily associated with the closure of heart valves, offer crucial insights into cardiac function. The first heart sound (S1), produced by the closure of the atrioventricular valves, and the second heart sound (S2), associated with the closure of the semilunar valves, are commonly recognized as the characteristic ’lub’ and ’dub.’ In a healthy heart, these sounds provide clear indications of proper valve function. Over time, the practice of analyzing these sounds has become an indispensable non-invasive tool in medical diagnostics, offering a reliable means to detect abnormalities and evaluate cardiac performance. I-A Anomalies in the Heart Sounds Heart murmurs, extra heart sounds, and extrasystoles are common anomalies detected during cardiac auscultation. Each type of anomaly provides valuable information about potential underlying cardiac conditions. Heart murmurs are among the most common anomalies detected through auscultation. Produced by turbulent blood flow strong enough to generate audible noise, these ”whooshing” sounds can be heard in various scenarios, including some healthy individuals. While innocent murmurs, also known as functional or benign murmurs, are typically harmless and not associated with structural heart abnormalities, pathologic murmurs indicate underlying conditions such as valve defects, congenital heart defects, or abnormal blood flow patterns. Extra Heart Sounds refer to additional heart sounds beyond the normal ”lub-dub” pattern. These may manifest as ”lub-lub dub” or ”lub dub-dub” sequences. Extra heart sounds can sometimes indicate underlying conditions, although they may also occur in healthy individuals. Detection of these sounds is crucial as they may not be easily identified through other diagnostic tools like ultrasound. Extrasystole involves irregular heart rhythms, typically presenting as extra or skipped heartbeats. These can be heard as sequences such as ”lub-lub dub” or ”lub dub-dub.” While extrasystoles may be benign, they can also signify underlying heart disease, making early detection important for effective treatment. A detailed explanation of these anomalies can be found in Table I . I-B Organization of the Paper This paper is organized as follows: Section II covers the background of the study, providing the necessary theoretical foundation. III provides an in-depth review of related work, discussing the advancements and methodologies in cardiac sound analysis and the integration of machine learning models in cardiovascular diagnostics. Section IV details the proposed approach, including the data preprocessing techniques, the generation of audiovisual data, and the model architectures. Section V describes the experimental setup, including the dataset used, training procedures, and evaluation metrics. The results of the experiments are presented in VI , highlighting the performance of individual models and the ensemble method. Finally, Section VII concludes the paper with a summary of findings, potential implications for clinical practice, and directions for future research. I-C Contributions The primary impacts of the proposed experiment are the following: • We developed ENACT-Heart (See Fig. ) – a novel transformer-based ensemble method specifically designed for diagnosing medical audio data through advanced visualization techniques. ENACT-Heart demonstrates state-of-the-art performance, surpassing other existing ensemble methods in the field. • We explored the feasibility of employing the Mixture of Experts (MoE) approach across different AI architectures, combining CNN and ViT. This integration effectively leverages multiple image modalities derived from the same input data, enhancing diagnostic accuracy. • We applied a ViT model to medical time-series sound data by converting it into audiovisual representations. This innovative approach, still an active area of research, opens new avenues for analyzing and interpreting complex medical signals. TABLE I: Types of anomalies in heart sound [ 1 ] . Category Description Normal • Healthy heart sounds with a clear ”lub dub” pattern. • May contain background noises and occasional random noise. Murmur • Abnormal heart sounds with a ”whooshing, roaring, rumbling, or turbulent fluid” noise between ”lub” and ”dub”, or between ”dub” and ”lub”. • ”Lub” and ”dub” are still present. • Murmurs do not occur directly on ”lub” or ”dub”. Extra Heart Sound • Additional heart sounds such as ”lub-lub dub” or ”lub dub-dub”. • May or may not be a sign of disease. • Important to detect as it may not be detected well by ultrasound. Extrasystole • Heart sounds out of rhythm involving extra or skipped heartbeats, such as ”lub-lub dub” or ”lub dub-dub”. • May or may not be a sign of disease. • Treatment is likely to be more effective if diseases are detected earlier. II Historical Background The diagnosis of heart conditions dates back to the early days of medicine, where physicians relied on palpation and pulse assessment to detect abnormalities. A significant breakthrough occurred in the 1700s when Jean Baptiste de Senac, physician to King Louis XV of France, established the connection between atrial fibrillation and mitral valve disease. Senac’s work laid the foundation for cardiology as a distinct field of study [ 2 ] . The invention of the stethoscope by René Laennec in 1816 marked a pivotal moment in cardiac diagnostics. Laennec introduced the technique of ”mediate auscultation” using his newly created paper acoustic device, allowing for more accurate detection of heart sounds and abnormalities [ 3 ] . This innovation remains a cornerstone in the history of cardiology. These early diagnostic methods, based on manual interpretation of heart sounds, evolved significantly over the centuries. With technological advancements, traditional auscultation has been augmented by electrocardiograms (ECGs) and other imaging modalities. In recent years, the integration of artificial intelligence (AI) and machine learning has opened new avenues for the analysis of heart sounds, leading to more precise and efficient diagnostic tools [ 4 , 5 ] . III Related Works III-A Thematically Related Works Thematically, the proposed method lies in the field of computer-assisted diagnosis (CAD) systems for heart disease. CAD systems for heart diseases utilize computation techniques such as machine learning, pattern recognition, and AI to analyze cardiac data and provide decision-support tools for healthcare providers. There have been numerous attempts to apply CAD systems for heart diseases in diverse modalities, including but not limited to ECG, cardiac CT/MRI, etc. These systems analyze cardiac data to identify abnormalities and patterns - especially indicators of certain heart diseases. However, the direction of the majority of these researches are pointed mainly toward computer vision over audio AI, mainly due to the advanced deep learning models available. For some researches that emphasized sound classification, its methodologies have varied slightly from the approach proposed in this study. For instance, Jumphoo et al. utilized a CNN for feature extraction and Data-efficient Image Transformer (DieT), a variant of the

Read the complete text