AI Meets Motherhood: Transforming Neonatal Health Monitoring with Contactless Physiological Analysis
Neonatal health monitoring is essential for safeguarding the well-being of newborns. Traditional contact-based monitoring methods, while effective, are often inconvenient and unsuitable for continuous use in neonates with sensitive skin, limited scalability, and insufficient attention to dynamic, long-term physiological monitoring. Emerging advancements in tele-healthcare, IoT, and remote photoplethysmography (rPPG) present an opportunity to revolutionize monitoring through non-contact, scalable, and transparent approaches. rPPG leverages facial video to estimate physiological parameters such as heart rate, respiratory rate, blood pressure and stress levels, offering a promising alternative to contact-based sensors. However, existing rPPG methods struggle under challenging conditions such as neonatal crying, occlusions, and variable illumination. Additionally, current deep learning-based rPPG models lack explainability, impeding trust and adoption in clinical settings.
To overcome these challenges, we will develop an innovative rPPG system tailored for neonates. We propose a hybrid approach integrating Eulerian and Lagrangian techniques to enhance robustness against motion artifacts and occlusions. Dynamic region-of-interest selection and signal fusion will improve data quality under challenging conditions. The project will also leverage advanced deep learning models, including transformer-based architectures, to optimize signal extraction and denoising. A critical focus will be on neonates, investigating the optimal use of blue color channels for rPPG signals due to their thinner skin and unique optical properties. Explainable AI will ensure transparency in predictions, fostering trust among clinicians, caregivers, and stakeholders. Data collection and pilot studies will be conducted in collaboration with local hospitals, encompassing diverse behavioral and environmental scenarios.
This research promises transformative impact by providing scalable, cost-effective, and accessible healthcare solutions for underserved populations. It aims to advance neonatal care through continuous, contactless monitoring, enabling early detection of complications, improved outcomes, and enhanced trust in AI-driven healthcare solutions
Key words:
Health monitoring, deep learning, explainable AI, tele-healthcare, remote photoplethysmography (rPPG)
AI for heart monitoring using exercise aware wearables (Exercise4Heart)
Aims:
Heart failure impacts 15 million individuals in Europe and accounts for 2% of healthcare expenses in the European Union. The Exercise4Heart initiative seeks to create a continuous monitoring system utilizing wearable technology for early detection of worsening heart failure.
Methods:
The commercialized invasive solution for early detection of heart failure (HF) has shown the feasibility of early detection of HF but in an invasive way with multiple disadvantages. Current non-invasive methods for early detection of HF typically rely on analyzing electrocardiograms (ECG) taken during periods of rest. However, there is a gap in the literature concerning the simultaneous utilization of electrocardiography (ECG) and seismocardiography (SCG) data collected during periods of both rest and physical exercise to detect heart failure (HF). Building upon the effective application of a comparable methodology in the detection of COVID-19 and recent studies suggesting the potential of SCG for HF detection, as well as the close correlation between exercise intolerance and HF diagnosis, there exists the possibility of early detection of HF deterioration using this combined method.
This study aims to employ semi-supervised and memory-efficient artificial intelligence (AI) models to analyze ECG and SCG data collected during both rest and physical exercise, such as walking, to enable early HF identification. To develop a memory-efficient solution, data scaling techniques will be utilized to condense the raw ECG and SCG signals, followed by feature extraction and AI model development for HF detection.