Time Series Recognition with Convolutional and Recursive Neural Networks in BSPM

• Published in: 2023 International Interdisciplinary PhD Workshop (IIPhDW) pp. 1 - 6
• Main Authors: Wojcik, Dariusz, Rymarczyk, Tomasz, Maciura, Lukasz, Oleszek, Michal, Adamkiewicz, Przemyslaw
• Format: Conference Proceeding
• Language: English
• Published: IEEE 03.05.2023
• Subjects: Body Surface potential mapping; Convolution; convolutional and recurrent neural networks; Electrocardiography; Heart; Logic gates; Machine learning; Recurrent neural networks; time series; Time series analysis
• DOI: 10.1109/IIPhDW54739.2023.10124427

This paper presents biomedical time series from Body Surface Potential Mapping (BSPM) recognition using various convolutional and recurrent neural network structures. The BSPM signal is in a form of time series form 102 channels located around the chest. The time series are then transformed to time windows to allow recognition of heart diseases. The several options of neural network structures were compared: one-dimensional convolutional neural network, Long-Short-Term Memory neural network, and Gated Recurrent Unit neural network. The article showcases different convolutional and recurrent neural network architectures for recognizing patterns in biomedical time series measured with Body Surface Potential Mapping. The study compared three types of neural network structures: Long-Short-Term Memory neural network, Gated Recurrent Unit neural network, and one-dimensional convolutional neural network. The main goal of paper is to find optimal machine learning solution for heart disease recognition on the basis of BSPM signal. The best results are obtained using model with GRU layer.