Scholarlink

<p>Bidirectional Long short-term memory (LSTM) units have recently emerged as a boon in the analysis of time-series data. LSTM are a type of Recurrent neural network. In this study, a new DeepBiLSTMnet architecture is proposed that facilitates detection of cardiac arrhythmia. Using this detected arrhythmia beats an inference engine is designed which predicts the severity of illness. The methodology begins with collection of ECG data from MIT-BIH database. The ECG data are then pre-processed. Then, the proposed DeepBiLSTMnet architecture is designed using a wavelet sequence layer and Bi-LSTM layer followed by classification layer with SoftMax activation. Bi-LSTM layer is designed by sequentially connecting 200 hidden Bi-LSTM units. The model is then trained with different network training parameter configurations. Investigating the obtained results by the process of training, the model with system's best training accuracy is selected and tested with test data. To update the weights and offsets, our model is tested with different optimizers. To develop a prototype NVIDIA Jetson Nano Developer Kit is used. Our proposed model gave a high recognition performance with a model accuracy of 99.59%. Other than overall model accuracy, various performance metrics like Precision, Specificity, Recall, F-score and Class accuracy for each class are also calculated. Furthermore, the Inference engine is designed using the obtained classification results to predicted the severity of illness and validated with real-time data.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A novel inference system for detecting cardiac arrhythmia using deep learning framework

  • 期刊论文

  • 发表日期：

    2025年3月7日

查看全文
  • Mohebbanaaz,
  • Y. Padma Sai,
  • L. V. Rajani Kumari

摘要

Bidirectional Long short-term memory (LSTM) units have recently emerged as a boon in the analysis of time-series data. LSTM are a type of Recurrent neural network. In this study, a new DeepBiLSTMnet architecture is proposed that facilitates detection of cardiac arrhythmia. Using this detected arrhythmia beats an inference engine is designed which predicts the severity of illness. The methodology begins with collection of ECG data from MIT-BIH database. The ECG data are then pre-processed. Then, the proposed DeepBiLSTMnet architecture is designed using a wavelet sequence layer and Bi-LSTM layer followed by classification layer with SoftMax activation. Bi-LSTM layer is designed by sequentially connecting 200 hidden Bi-LSTM units. The model is then trained with different network training parameter configurations. Investigating the obtained results by the process of training, the model with system's best training accuracy is selected and tested with test data. To update the weights and offsets, our model is tested with different optimizers. To develop a prototype NVIDIA Jetson Nano Developer Kit is used. Our proposed model gave a high recognition performance with a model accuracy of 99.59%. Other than overall model accuracy, various performance metrics like Precision, Specificity, Recall, F-score and Class accuracy for each class are also calculated. Furthermore, the Inference engine is designed using the obtained classification results to predicted the severity of illness and validated with real-time data.