Heart Sound Classification Using Deep Learning Techniques Based on Log-mel Spectrogram

Nguyen, Minh-Tuan; Lin, Wei; Huang, Jin

doi:10.1007/s00034-022-02124-1

Cited by 28 publications

(23 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Section 1.3). However, we argue that the findings from existing studies [8,9,81] are not directly comparable to the 2022 Challenge. For example, the Yaseen dataset [30]-often cited for models with accuracy rates exceeding 99%-features extremely short (<4 sec) and clean recordings.…”

Section: Outcome Task Performancementioning

confidence: 58%

From Theoretical Models to Practical Deployment: A Perspective and Case Study of Opportunities and Challenges in AI-driven Cardiac Auscultation Research for Low-Income Settings

Krones,

Walker

2023

Preprint

View full text Add to dashboard Cite

This paper critically explores the opportunities and challenges of deploying Artificial Intelligence (AI) in healthcare. This study has two parallel components: (1) A narrative literature summary, which assesses the capacity of AI to aid in addressing the observed disparity in healthcare between high- and low-income countries. Despite the development of machine learning models for a wide range of diseases, many are never deployed in practice. We highlight various challenges that contribute to the lack of deployed models. A main challenge that is not always sufficiently addressed in the literature is the evaluation of model generalisation. For example, by using a multi-site set-up with test sets that were collected separately to the train and validation sets, or by using evaluation metrics which are both understandable and clinically applicable. Moreover, we discuss how the emerging trend of human-centred deployment research is a promising avenue for overcoming barriers towards deployment. (2) A case study on developing and evaluating a predictive AI model tailored for low-income environments. The focus of this case study is heart murmur detection in rural Brazil. Our Binary Bayesian ResNet model leverages overlapping log mel spectrograms of patient heart sound recordings and integrates demographic data and signal features via XGBoost to optimise performance. We discuss the model’s limitations, its robustness, and the obstacles preventing its practical application. We especially highlight how our model, and other state-of-the-art models, struggle to generalise to out-of-distribution data. The research accentuates the transformative potential of AI-enabled healthcare, particularly affordable point-of-care monitoring systems, in low-income settings. It also emphasises the necessity for effective implementation and integration strategies to guarantee the successful deployment of these technologies.

show abstract

Section: Outcome Task Performancementioning

confidence: 58%

From Theoretical Models to Practical Deployment: A Perspective and Case Study of Opportunities and Challenges in AI-driven Cardiac Auscultation Research for Low-Income Settings

Krones,

Walker

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Commonly used 2D representations of heartbeat signals include Mel spectrograms and short‐time Fourier transform time‐frequency maps. In the literature (Nguyen et al, 2023), two DL models, the LSTM and the 2D CNN, are suggested for categorizing heartbeat sounds based on extracted features. The log‐mel spectrum characteristics are then retrieved after the cardiac signal has been initially cropped to a constant length.…”

Section: Deep Learning For Heart Sound Classificationmentioning

confidence: 99%

Artificial intelligence for heart sound classification: A review

Chen,

Guo,

et al. 2024

Expert Systems

View full text Add to dashboard Cite

Heart sound signal analysis is very important for the early identification and treatment of cardiovascular illness. With rapid advancements in science and technology, artificial intelligence technologies are providing tremendous opportunities to enhance diagnosis and clinical decision‐making. Instruments can now perform clinical diagnoses that previously could only be handled by human experts more conveniently and efficiently. Despite multiple works on automatic heart sound analysis, there are few summarization and review works. This article attempts to give a thorough overview of various heart sound analysis subtasks and examine the improvements made in each subtask by both machine learning techniques and deep learning algorithms. It goals to highlight the potential of AI to revolutionize cardiovascular healthcare by enabling accurate and automated analysis of heart sounds. The findings of this review are beneficial for researchers, clinicians, and engineers in the development and application of AI‐based solutions for improved heart sound classification and diagnosis.

show abstract

“…Real numbers to mel spectrograms: First, the signal is resampled from 1000 to 44 100 Hz and then passed on to a function to convert it to a mel spectrogram. 38 Later, its power is converted to decibels and then plotted and saved as a mel spectrogram as shown in Figure 4. Let x(t) be the time-domain EEG signal, FFT(x(t)) be the fast Fourier transform of the signal, and P x (f ) be the power spectrum at frequency f of the EEG signals divided into window size W. So, the power spectrum can be calculated as follows,…”

Section: Bad Channel(𝜒mentioning

confidence: 99%

“…

\mathrm{EEG}\ \mathrm{signal}\overset{\mathrm{C} WT}{\to \limits}\mathrm{Wavelets}.

Real numbers to mel spectrograms: First, the signal is resampled from 1000 to 44 100 Hz and then passed on to a function to convert it to a mel spectrogram 38 . Later, its power is converted to decibels and then plotted and saved as a mel spectrogram as shown in Figure 4.Let

x(t)

be the time‐domain EEG signal,

\mathrm{FFT}\left(x(t)\right)

be the fast Fourier transform of the signal, and

{P}_x(f)

be the power spectrum at frequency

f

of the EEG signals divided into window size

W

.…”

Section: Proposed Architecturementioning

confidence: 99%

See 1 more Smart Citation

EEG‐based biometric authentication system using convolutional neural network for military applications

Vadher,

Patel,

Nair

et al. 2023

Security and Privacy

View full text Add to dashboard Cite

In this technological era, as the need for security arises, the use of biometrics is increasing in authentication systems as a secure and convenient method of human identification and verification. Electroencephalogram (EEG) signals have gained significant attention among the various biometric modalities available because of their unique and unforgeable characteristics. In this study, we have proposed an EEG‐based multi‐subject and multi‐task biometric authentication system for the military applications that address the challenges associated with multi‐task variation in EEG signals. The proposed work considers the use of respective EEG signals for the access of artillery, entrance to highly confidential places for the military and so forth by authenticated personnel only. We have used a multi‐subject, multi‐session, and multi‐task () dataset. The dataset was partially preprocessed with basic signal processing techniques such as bad channel repairing, independent component analysis for artifact removal, downsampling to 250 Hz, and an audio filter of 0.01–200 Hz for signal improvisation. This partially preprocessed dataset was further processed and was used in our deep learning model (DL) architectures. For EEG‐based biometric authentication, convolutional neural network (CNN) outperforms many of the state‐of‐the‐art DL architectures with a validation accuracy of approximately 99.86%, training accuracy of 98.49% and precision, recall and F1‐score with values of 99.91% that makes this EEG‐based approach for authentication more reliable. The DL models were also compared based on training and inference time, where CNN used the most training time but took the least time to predict the output. We compared the performance of the CNN model for three preprocessing techniques by feeding mel spectrograms, chromagrams and mel frequency cepstral coefficients, out of which mel spectrograms provided better results. This proposed architecture proves to be robust and efficient for military applications.

show abstract

Heart Sound Classification Using Deep Learning Techniques Based on Log-mel Spectrogram

Cited by 28 publications

References 20 publications

From Theoretical Models to Practical Deployment: A Perspective and Case Study of Opportunities and Challenges in AI-driven Cardiac Auscultation Research for Low-Income Settings

From Theoretical Models to Practical Deployment: A Perspective and Case Study of Opportunities and Challenges in AI-driven Cardiac Auscultation Research for Low-Income Settings

Artificial intelligence for heart sound classification: A review

EEG‐based biometric authentication system using convolutional neural network for military applications

Contact Info

Product

Resources

About