Multitask Siamese Network for Remote Photoplethysmography and Respiration Estimation.

Area of Science:

• Biomedical Engineering
• Computer Vision
• Machine Learning

Background:

• Vital signs like heart and respiration rates are crucial for health assessment.
• Accurate and accessible vital sign monitoring is essential for remote patient care and personal health tracking.

Purpose of the Study:

• To propose a novel multitask Siamese network (MTS) model for accurate, simultaneous prediction of heart and respiration rates.
• To develop a lightweight model suitable for implementation on edge devices like mobile phones.

Main Methods:

• Utilized a multitask Siamese network architecture trained on facial images (cheek, nose, mouth, forehead).
• Employed shared parameters between Siamese networks to extract cardiac and respiratory information.
• Compared MTS model performance against single-task learning and conventional multitask learning models.

Main Results:

• The MTS model achieved high vital sign prediction accuracy comparable to single-task models.
• Outperformed conventional multitask learning models in accuracy.
• Reduced model parameters by 16 times, with mean average errors of 2.84 for heart rate and 4.21 for respiration rate.

Conclusions:

• The proposed MTS model enables simultaneous prediction of heart and respiration rates with high accuracy and reduced parameters.
• Its lightweight nature makes it highly suitable for real-time vital sign monitoring on edge devices.
• Offers a promising solution for accessible and efficient health monitoring.