Phonocardiography Signals Compression with Deep Convolutional Autoencoder for Telecare Applications

Chien, Ying-Ren; Hsu, Kai-Chieh; Tsao, Hen‐Wai

doi:10.3390/app10175842

Cited by 10 publications

(5 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, by open-sourcing the Biosignal Data Compression Toolbox, we expect to cultivate a community where researchers will test and contribute novel data compression methods on the data that we have released, further expanding the reach of the current study. Recent research in using deep convolutional autoencoders for biosignal data compression [ 35 , 36 ] has shown great promise and should be included in the Biosignal Data Compression Toolbox in future work. Future work will involve developing and testing methods for quantifying signal characteristics to determine optimal data compression pipelines for other biosignals.…”

Section: Discussionmentioning

confidence: 99%

Biosignal Compression Toolbox for Digital Biomarker Discovery

Bent

Kim

et al. 2021

Sensors

View full text Add to dashboard Cite

A critical challenge to using longitudinal wearable sensor biosignal data for healthcare applications and digital biomarker development is the exacerbation of the healthcare “data deluge,” leading to new data storage and organization challenges and costs. Data aggregation, sampling rate minimization, and effective data compression are all methods for consolidating wearable sensor data to reduce data volumes. There has been limited research on appropriate, effective, and efficient data compression methods for biosignal data. Here, we examine the application of different data compression pipelines built using combinations of algorithmic- and encoding-based methods to biosignal data from wearable sensors and explore how these implementations affect data recoverability and storage footprint. Algorithmic methods tested include singular value decomposition, the discrete cosine transform, and the biorthogonal discrete wavelet transform. Encoding methods tested include run-length encoding and Huffman encoding. We apply these methods to common wearable sensor data, including electrocardiogram (ECG), photoplethysmography (PPG), accelerometry, electrodermal activity (EDA), and skin temperature measurements. Of the methods examined in this study and in line with the characteristics of the different data types, we recommend direct data compression with Huffman encoding for ECG, and PPG, singular value decomposition with Huffman encoding for EDA and accelerometry, and the biorthogonal discrete wavelet transform with Huffman encoding for skin temperature to maximize data recoverability after compression. We also report the best methods for maximizing the compression ratio. Finally, we develop and document open-source code and data for each compression method tested here, which can be accessed through the Digital Biomarker Discovery Pipeline as the “Biosignal Data Compression Toolbox,” an open-source, accessible software platform for compressing biosignal data.

show abstract

Section: Discussionmentioning

confidence: 99%

Biosignal Compression Toolbox for Digital Biomarker Discovery

Bent

Kim

et al. 2021

Sensors

View full text Add to dashboard Cite

show abstract

“…Using machine learning and clustering algorithms, this study classified and localized heartbeats using phonocardiogram signals [28]. The K-means clustering algorithm, which is based on the amplitude of the signal, generated clusters.…”

Section: B Artificial Intelligence Based Methodsmentioning

confidence: 99%

Adopting Artificial Intelligence Algorithms for Remote Fetal Heart Rate Monitoring and Classification using Wearable Fetal Phonocardiography

ABBURI¹,

Hatai²,

Samanta³

2023

Preprint

View full text Add to dashboard Cite

<p>This work utilizes a wearable phonocardiogram belt for acquiring signals from remote areas. Four sequential methodologies are undergone in the proposed work for resilience, and effectiveness. Intelligent, and accurate FHR monitoring. At first, the noise and artifacts are removed by performing two levels pre-processing methodology method. The low frequency noises are removed by Chebyshev II High Pass Filter and high frequency noises are removed by a hybrid of EC2EMDAN- PS-MODWT filters respectively. In the next step, the denoised signals are segmented for reducing the complexity in which the segmentation is performed using Multi Agent Deep Q-Learning (MA-DQL) algorithm based on several constraints. Further, the segmented signal is provided for reducing the redundancies in cardiac cycles using Artificial Humming Bird Optimization (AHBO) algorithm. The segmented and non-redundant signals are converted into 3D spectrograms using a machine learning algorithm named Variational Auto Encoder-General Adversarial Networks (VAE-GAN) for analyzing the signals with better visual interpretation in 3D space. Finally, utilizing the 3D spectrogram analysis the feature extraction and classification is taken place by adopting a hybrid of Bidirectional Gated Recurrent Unit (BiGRU) and Multi Boosted Capsule Network (MBCapsNet) into three classes such as normal (110-160 BPM), abnormal (above 180 BPM), and Suspicious (fluctuates between normal and abnormal). The proposed work is implemented and simulated in the MATLAB R2020a tool and the performance is validated by adopting effective validation metrics. The outcomes demonstrate that the suggested work performs better than the current works. </p>

show abstract

“…They are expanded to 8 seconds by padding zeros to the end of the records. The zero-padding has been used for PCG data with CNN in the literature (Noman et al 2019, Chien et al 2020. The fragmented recordings from PASCAL dataset were mainly from Dataset B (95% of the recordings) as most of the recordings in Dataset A are less than 8 seconds.…”

Section: Dataset Preparationmentioning

confidence: 99%

Deep learning based classification of unsegmented phonocardiogram spectrograms leveraging transfer learning

Khan

Abid

et al. 2021

Physiol. Meas.

View full text Add to dashboard Cite

Cardiovascular diseases (CVDs) are the main cause of deaths all over the world. Heart murmurs are the most common abnormalities detected during the auscultation process. The two widely used publicly available phonocardiogram (PCG) datasets are from the PhysioNet/CinC (2016) and PASCAL (2011) challenges. The datasets are significantly different in terms of the tools used for data acquisition, clinical protocols, digital storages and signal qualities, making it challenging to process and analyze. In this work, we have used short-time Fourier transform (STFT) based spectrograms to learn the representative patterns of the normal and abnormal PCG signals. Spectrograms generated from both the datasets are utilized to perform three different studies: (i) train, validate and test different variants of convolutional neural network (CNN) models with PhysioNet dataset, (ii) train, validate and test the best performing CNN structure on combined PhysioNet-PASCAL dataset and (iii) finally, transfer learning technique is employed to train the best performing pre-trained network from the first study with PASCAL dataset. We propose a novel, less complex and relatively light custom CNN model for the classification of PhysioNet, combined and PASCAL datasets. The first study achieves an accuracy, sensitivity, specificity, precision and F1 score of 95.4%, 96.3%, 92.4%, 97.6% and 96.98% respectively while the second study shows accuracy, sensitivity, specificity, precision and F1 score of 94.2%, 95.5%, 90.3%, 96.8% and 96.1% respectively. Finally, the third study shows a precision of 98.29% on the noisy PASCAL dataset with transfer learning approach. All the three proposed approaches outperform most of the recent competing studies by achieving comparatively high classification accuracy and precision, which make them suitable for screening CVDs using PCG signals.

show abstract

Phonocardiography Signals Compression with Deep Convolutional Autoencoder for Telecare Applications

Cited by 10 publications

References 43 publications

Biosignal Compression Toolbox for Digital Biomarker Discovery

Biosignal Compression Toolbox for Digital Biomarker Discovery

Adopting Artificial Intelligence Algorithms for Remote Fetal Heart Rate Monitoring and Classification using Wearable Fetal Phonocardiography

Deep learning based classification of unsegmented phonocardiogram spectrograms leveraging transfer learning

Contact Info

Product

Resources

About