Skip to main navigation Skip to search Skip to main content

PhiVitalNet: A Physics-Informed Deep Learning for Contactless Vital Signs Monitoring

  • Ridwanullahi Isa*
  • , Abdullah M. Algarni
  • , Abdulmajid Lawal
  • , Rifaqat Hussain
  • *Corresponding author for this work

Research output: Contribution to journal › Article › peer-review

Abstract

Continuous monitoring of human vital signs is essential for timely detection of physiological abnormalities in clinical settings. Conventional contact-based respiration belts and electrocardiograms restrict patient movement, whereas contactless radar-based sensing enables comfortable long-term physiological monitoring. However, raw radar data contain physiological signals mixed with harmonics and motion artifacts, making it difficult to extract respiratory and cardiac components accurately. To address these challenges, this study proposes a physics-informed deep-learning framework that leverages a convolutional neural network (CNN) and transformer model. The proposed approach combines spectral feature extraction via a CNN with transformer attention mechanisms to estimate the respiration rate (RR) and heart rate (HR) and to reconstruct respiratory and cardiac spectra. This study used a single model to estimate both RR and HR simultaneously, thereby reducing overhead. In addition, the physics-guided framework combines spectral-domain learning with differentiable spectral peak alignment loss within established bands to improve detection accuracy. Evaluations under both mild and noisy conditions indicate that the proposed framework achieves improved cross-subject generalization over classical techniques and standalone deep learning baselines, with mean absolute errors (MAEs) of 1.13 breaths per minute for RR and 2.71 beats per minute for HR.

Original languageEnglish
Pages (from-to)154557-154569
Number of pages13
JournalIEEE Access
Volume14
DOIs
StatePublished - 2026

Bibliographical note

Publisher Copyright:
© 2013 IEEE.

Keywords

  • Contactless sensing
  • convolutional network
  • physics-informed
  • radar-based
  • vital signs

ASJC Scopus subject areas

  • General Computer Science
  • General Materials Science
  • General Engineering

Fingerprint

Dive into the research topics of 'PhiVitalNet: A Physics-Informed Deep Learning for Contactless Vital Signs Monitoring'. Together they form a unique fingerprint.

Cite this