Smartphone-based health monitoring in India: Data collection and evaluation for pulse rate estimation

ABSTRACT Introduction: Over the past decade, monitoring of body vitals has gained significant popularity, specifically during and post the recent COVID pandemic. Advancements in smartphones and wearables have been pivotal, providing accessible and cost-effective solutions for at-home health monitoring. Their development often requires a large corpus of labeled datasets, but such large and diverse datasets for developing smartphone-based vital estimation systems, particularly adapted to Indian context, are scarce. Aims and Objectives: This observational study focuses on development of such a dataset in a diverse Indian context and evaluation of smartphone-based pulse rate estimation based on this dataset. Methods: Data collection considered Indian patients with various medical conditions, body mass index profiles, blood pressure levels, ages, and smoking habits, reflecting a broad demographic spectrum. As part of this study, an algorithm was implemented to estimate the photoplethysmogram (PPG) signal from video recordings of fingers placed on the smartphone camera and subsequently to estimate pulse rate using the acquired PPG data. Smartphone-based pulse rate estimates were compared with readings from pulse oximeters to assess accuracy and feasibility. Results: The smartphone-based PPG algorithm provides reasonably accurate estimations of pulse rate when compared to traditional pulse oximeters under varied healthcare settings (mean absolute error 0.90). Conclusion: Results indicate that the smartphone-based PPG signal captures sufficient information of the cardiac cycle to reliably estimate the pulse rate. Furthermore, system accuracy is consistent across varied subjects and settings, highlighting the importance of tailored data collection for development and evaluation of vital estimation algorithms. Keywords: Machine learning, oximeter, photoplethysmogram, pulse rate, smartphone