A smartphone-based cardiac monitoring app using computer vision for hypertension management

Chong, Jia Hui (2026) A smartphone-based cardiac monitoring app using computer vision for hypertension management. Final Year Project, UTAR.

PDF Download (10Mb)

Abstract

Hypertension is a major risk factor for cardiovascular disease and it is often unaware due to it shows no symptoms. This project proposes a smartphone-based cardiac monitoring application to support hypertension self-management and awareness. The application uses smartphone camera-based finger photoplethysmography (PPG) to estimate heart rate (HR), heart rate variability (HRV), and blood pressure (BP). A real-time computer vision technique, the red, green and blue (RGB) light signal extraction method is combined with a robust PPG signal quality assessment module to ensure only reliable signals are processed. The application supports manual BP entry, automatic health classification such as low, normal, elevated or high, trend analysis and personalized recommendations. Other features such as the chatbot-based AI Cardio Assistant, reminders and health record management are also provided. This project shows that the signal quality model using Logistic Regression with physiological check achieved 81% accuracy and 92% precision. The BP estimation model using XGBoost with Ridge Regression calibration meets the ISO, AAMI, and BHS standards, indicating strong potential for cuff-less BP monitoring. Explainable Artificial Intelligence (XAI), including Permutation Importance and Tree SHapley Additive exPlanations (TreeSHAP) further confirms that the BP predictions are driven by meaningful physiological features rather than irrelevant factors. HR estimation using the HeartPy algorithm shows good agreement with the reference measurements, while HRV shows low correlation with ECG and is therefore used only for trend indication. The models are fully integrated into a mobile application developed using Flutter. Firebase is used for authentication and data storage, while Render with FastAPI is used for hosting the machine learning models. In conclusion, the proposed application demonstrates a functional smartphone-based cardiac monitoring application for preventive hypertension management. Future work will focus on improving the BP estimation using a more diverse dataset, enhancing the real-time performance and integrating location-based navigation features in the application.

Actions (login required)