Title: MED-NET: a novel approach to ECG anomaly detection using LSTM auto-encoders
Authors: Koustav Dutta; Rasmita Lenka; Soumya Ranjan Nayak; Asimananda Khandual; Akash Kumar Bhoi
Addresses: Kalinga Institute of Industrial Technology, Deemed to be University, Bhubaneswar, Odisha, India ' Kalinga Institute of Industrial Technology, Deemed to be University, Bhubaneswar, Odisha, India ' Amity School of Engineering and Technology, Amity University Uttar Pradesh, Noida, India ' Department of Textile, College of Engineering and Technology, Bhubaneswar, Odisha, India ' Department of Electrical & Electronics Engineering, Sikkim Manipal Institute of Technology (SMIT), Sikkim Manipal University (SMU), Majitar, Sikkim, India
Abstract: Time series data is generated in various sectors of day to day life. Among all, one of the most important areas of generation and processing of time series data plays vital role in medical analysis. In this context, various continuous time series dependent EEG and ECG (electrocardiogram) signals are the most important types of medical signals produced and monitored by doctors. This paper proposes a highly novel and robust approach to analyse and detect ECG signals for tracking of anomalies in the signals using Hybrid Deep Learning Architectures (HDLA). The proposed scheme implements self-supervised pattern recognition using Long Short-Term Memory (LSTM) networks in terms of autoencoder and decoder. Finally, the proposed scheme is tested on Physio-net data set. The outcome of the model can also handle noise associated with ECG time series signal, it achieves high accuracy and also solves overfitting problems in a robust and efficient manner.
Keywords: bio-signals; encoder; decoder; LSTM; auto-encoder; ECG; anomaly; time series; hybrid model; reconstruction error.
DOI: 10.1504/IJCAT.2021.117277
International Journal of Computer Applications in Technology, 2021 Vol.65 No.4, pp.343 - 357
Received: 06 Jul 2020
Accepted: 17 Aug 2020
Published online: 31 Aug 2021 *