Volume 44 Issue 1
Apr.  2024
Turn off MathJax
Article Contents

SEGMENTATION AND CLASSIFICATION OF CARDIAC BEATS USING BIDIRECTIONAL LONG SHORT-TERM MEMORY NETWORK FRAMEWORK

  • In the realm of automated heart disease diagnosis, the segmentation of electrocardiogram (ECG) signals to distill meaningful features is pivotal for reducing complexity. This endeavor, crucial for dimensionality reduction, aims at improving the accuracy and speed of classification processes, which are vital in curbing mortality rates associated with cardiovascular issues. However, the detection process is riddled with challenges stemming from the nonstationarity and high variability inherent in ECG signals, complicating analysis in both time and frequency domains. Compounding these hurdles are imbalanced and indistinct datasets. In this study, we propose a novel approach utilizing deep learning, particularly recurrent neural networks with long short-term memory (LSTM) layers, to address dataset imbalances. LSTM networks excel at capturing sequential timing information inherent in ECG data. To counter dataset imbalances, we employ oversampling techniques and leverage focal loss-based weight balancing. This strategy significantly bolsters classification accuracy, with our proposed LSTM network achieving an impressive accuracy of 99.54%, surpassing traditional methods averaging around 98%.Furthermore, our approach demonstrates resilience to variations in ECG signal quality, thanks to an initial fuzzification process applied during dataset preprocessing. Looking ahead, the deployment of our method holds promise in bio-signal telemetry and pharmaceutical research, empowering physicians with robust tools to aid in diagnosis and treatment planning.

     

Catalog

    通讯作者: 陈斌, bchen63@163.com
    • 1. 

      沈阳化工大学材料科学与工程学院 沈阳 110142

    1. 本站搜索
    2. 百度学术搜索
    3. 万方数据库搜索
    4. CNKI搜索

    Figures(10)  / Tables(2)

    Article Metrics

    Article views (100) PDF downloads(50) Cited by()
    Proportional views
    Related

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return