Noninvasive Hemoglobin Quantification via Neural Network Optimization for Spectral Partialities

Le résumé fourni par la source

Anemia affects approximately one-quarter of the global population. While its diagnosis typically relies on hemoglobin (Hgb) measurement through venous blood draws, such methods require access to clinical infrastructure and trained personnel, limiting accessibility in rural and resource-limited settings. In addition, invasive blood sampling can cause pain, infection risk, and iatrogenic blood loss. Although noninvasive technologies have been developed, they often require expensive and bulky equipment, and their performance can be affected by skin tone variability across diverse populations. To overcome these limitations, we present a mobile health (mHealth) technology utilizing neural network models to noninvasively estimate blood Hgb levels based solely on red-green-blue (RGB) color values from smartphone photos of peripheral tissue. The palpebral conjunctiva (inner eyelid) is selected as a sensing site due to its easy accessibility, uniform microvasculature, and absence of confounding skin pigments. By optimizing neural networks with varying spectral partialities, we demonstrate that the proposed method can reliably predict blood Hgb levels using only limited RGB input, eliminating the need for complex and costly optical components. This study highlights how machine learning-powered mHealth technology can reduce dependence on clinical laboratories and hardware complexity, offering accessible and scalable solutions in resource-limited and home settings.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.