ECG Identification For Personal Authentication Using LSTM-Based Deep Recurrent Neural Networks | JoVE Visualize

Area of Science:

Biometrics and Security Engineering
Signal Processing and Machine Learning
Cardiovascular Health Monitoring

Background:

Traditional biometrics like fingerprint and face recognition face challenges such as forgery and environmental interference.
Electrocardiogram (ECG) signals offer a promising alternative for personal authentication, but conventional methods require long signal inputs, hindering real-time applications.
Existing ECG identification models, including RNN and DRNN, show potential but lack the efficiency needed for real-time biometric systems.

Purpose of the Study:

To develop a real-time, highly accurate biometric identification and classification system using ECG signals.
To address the limitations of existing ECG-based authentication methods, particularly the need for lengthy input signals and real-time processing.
To propose an advanced deep learning model for robust ECG signal analysis in biometric applications.

Main Methods:

A novel approach utilizing bidirectional long short-term memory (LSTM)-based deep recurrent neural networks (DRNN) with late-fusion for ECG biometric identification.
Implementation of a preprocessing pipeline including derivative filters, moving average filters, and normalization for efficient noise reduction and quick identification.
Experimental validation using two public datasets: MIT-BIH Normal Sinus Rhythm (NSRDB) and MIT-BIH Arrhythmia (MITDB).

Main Results:

The proposed LSTM-based DRNN model achieved exceptional performance on the NSRDB dataset, with 100% precision, recall, accuracy, and F1-score.
On the MITDB dataset, the model demonstrated high efficacy, achieving 99.8% precision, recall, accuracy, and a 0.99 F1-score.
The developed system exhibited superior classification accuracy and efficiency compared to conventional LSTM approaches for ECG-based biometrics.

Conclusions:

The proposed bidirectional LSTM-based DRNN model offers a highly accurate and efficient solution for real-time ECG-based biometric identification.
The preprocessing techniques enhance the model's ability to handle noisy ECG signals and enable rapid authentication.
This research advances the field of biometric security by providing a robust and practical ECG-authentication system suitable for real-world applications.