A Dataflow Framework for Cardiovascular Assessment Using Smartphones

Cardiovascular disease (CVD) prevention requires an integrated, patient-centered, and healthy-people-centered, interdisciplinary approach from several disciplines, and begins with CVD risk assessment and identification of specific risk conditions. The present study aims to develop a toolchain that contributes to preventive efforts in the early and presymptomatic diagnosis of cardiovascular diseases with the greatest public health impact. The framework discussed here comprises a web application with a web form for data acquisition, a moderated database with aggregate statistical records that are machine learning ready, a series of statistical and machine learning models, and a framework for model deployment. The main effort evolves from photoplethysmography (PPG), a noninvasive optical technology that measures changes in volumetric blood flow. Contrary to established techniques of using optical signals for PPG in the infrared part of the spectra, our ongoing research assesses the utilization of the visible part of the spectra. This allows wider practical application possibilities, as it can bring statistical and machine learning models to mobile devices, primarily smartphones. In this paper we present the conceptual framework of interdependent steps, starting from data collection for a specific tailored data science application, discussing challenges of data acquisition and validation, challenges of preparing machine-learning-ready dataset, and finally, model development. We discuss derived cardiovascular assessment parameters (e.g. heart-rate variability, perfusion index, respiratory rate, SpO2) and possibilities of blood pressure estimation from the PPG signal as a main goal. Research results may prove important for the efficient estimation of associated cardiovascular system risks.