CorNET: Deep Learning Framework for PPG-Based Heart Rate Estimation and Biometric Identification in Ambulant...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Miniaturized body-worn sensors enable pervasive biomedical signal monitoring.
• Remote cardiovascular monitoring benefits from non-invasive photoplethysmography (PPG) sensors in ambulatory settings.
• Wrist-worn PPG sensors are susceptible to motion artifacts, limiting their accuracy.

Purpose of the Study:

• To present a novel deep learning framework, CorNET, for estimating heart rate (HR) and performing biometric identification (BId).
• To utilize wrist-worn, single-channel PPG signals collected in an ambulant environment.
• To develop a personalized, data-driven approach for robust cardiovascular monitoring.

Main Methods:

• A four-layer deep neural network (CorNET) comprising two convolutional neural network (CNN) and two long short-term memory (LSTM) layers.
• A personalized, data-driven approach to model temporal sequences in PPG signals.
• Customized output layers for HR regression and BId classification.

Main Results:

• Achieved a mean absolute error of 1.47 ± 3.37 beats per minute for HR estimation.
• Attained an average accuracy of 96% for biometric identification on 20 subjects.
• Successfully validated CorNET in an ambulant use-case scenario with custom sensors.

Conclusions:

• CorNET demonstrates efficient and accurate HR estimation and BId using wrist-worn PPG signals in ambulant settings.
• The deep learning framework offers a promising solution for personalized remote cardiovascular monitoring.
• The proposed method effectively mitigates motion artifacts, enhancing the reliability of PPG-based health tracking.