Design and Validation of Ai-Enabled Pulse Monitoring Using Built-In Smartphone Sensors

Abstract

Smartphones have been transformed into formidable tools for real-time physiological monitoring as a result of the growing convergence of artificial intelligence (AI) and mobile health technology. Traditionally, dedicated medical or wearable devices have been used to measure pulse rate, which is an important indicator of cardiovascular health. However, widespread use of these devices is hindered by restrictions in terms of accessibility and cost. The objective of this research is to create, optimize, and evaluate an artificial intelligence-enabled pulse monitoring system that is capable of providing accurate, low-cost, and scalable cardiovascular evaluations. This will be accomplished by utilizing smartphone cameras and embedded sensors. The research draws on previous developments in artificial intelligence-driven biomedical sensing, digital cardiovascular modelling, and edge computing. It highlights the potential of lightweight machine learning architectures to extract clinically significant information from consumer-grade devices. The aims of the work include evaluating the accuracy of smartphone sensors, developing a lightweight CNN–LSTM model for pulse estimation, and testing the model's performance in comparison to medical-grade electrocardiogram (ECG) references across a wide range of environmental and demographic variables. With the help of 3,888 synchronized PPG–ECG recordings taken from the BUT-PPG dataset, a quantitative and experimental technique was successfully implemented. Among the several methods of signal processing were band-pass filtering, normalization, segmentation, and labeling based on electrocardiograms. TensorFlow-Lite was utilized in order to train and optimize the hybrid CNN–LSTM model for mobile deployment. A high prediction accuracy was observed in the results, with the mean absolute error (MAE) being around 1.96 beats per minute (BPM) and the critical correlation coefficient (CCC) being 0.896. Additionally, there was a substantial correlation with electrocardiogram data, and the performance was robust across different types of devices, lighting, movement, and measurement sites. According to the findings of the study, smartphones, when combined with AI architectures that have been optimized, are capable of providing near-clinical pulse monitoring without the need for external wearables. The significance of the system for scalable, accessible, and proactive cardiovascular health management is supported by the fact that it shows real-time performance, efficiency, and dependability that has been demonstrated.