Machine Learning Based Approach for Classification of Blood Flow using Non Invasive Device

Abstract

Abstract newlineCardiovascular diseases represent a significant global health challenge, necessitating accurate newlineand continuous monitoring of blood flow patterns for effective diagnosis and management. newlineTraditional methods for assessing blood flow, such as invasive catheterization and newlineDoppler ultrasound, present considerable limitations. Invasive procedures, while providing newlinehigh-fidelity data, are associated with patient discomfort, procedural risks, and the need for newlinespecialized medical infrastructure. Conversely, non-invasive techniques like Doppler newlineultrasound, though safer, often require skilled operators, can be uncomfortable for newlinecontinuous use, and may lack the sensitivity to detect subtle, early-stage abnormalities in newlineblood flow. This highlights a critical unmet need for an accessible, non-invasive, and reliable newlinetechnology that enables continuous, real-time monitoring of blood flow dynam- ics. Such newlinea device would not only enhance patient comfort and safety but also empower proactive and newlinepersonalized healthcare, facilitating the early detection of cardiovascular issues before they newlineescalate into more severe conditions. Addressing this gap is paramount to advancing newlinepreventative medicine and improving patient outcomes on a global scale. This research newlineintroduces an innovative, non-invasive device for the classification of blood flow patterns, newlineleveraging a novel dual-photoplethysmography (PPG) sensor configura- tion and an newlineoptimized machine learning architecture, termed PPG-Net 4. The thesis contributes by newlinedetailing custom-designed hardware that integrates two distinct low-cost PPG sensors, newlinestrategically placed on the fingertip (MAX32664) and wrist (MAX86150), to capture newlinecomprehensive PPG data. This dual-sensor approach, managed by a WiFi- enabled newlinemicrocontroller, allows the dataset to be transferred to a nearby local computer for analysis. newlineThe 10,000 mAh battery of this low-cost yet accurate hardware device allows data gathering newlinefor more than 24 hours. The core of the data analysis device is PPG-Net 4, a machine newlinelearning