Speaker
Description
Abstract. Background: Non-invasive hemoglobin (Hb) measurement has gained considerable interest as a painless alternative to blood sampling for anemia screening and continuous monitoring. Despite rapid advances in spectroscopy and optical sensing, clinical implementation remains limited by inconsistent diagnostic performance and heterogeneous validation methods.
Methods: This systematic review was conducted in accordance with the PRISMA 2020 guidelines. Literature published between 2015 and 2025 was identified through predefined search queries and citation chaining. Of 365 candidate studies, 344 were relevant, and 213 highly relevant studies were included in the qualitative synthesis. Data were extracted on sensing technology, optical principles, inference algorithms, confounder mitigation, validation methods, and diagnostic performance.
Results: Four major technology categories were identified: multi-wavelength photoplethysmography (37.6%), spectroscopy and hyperspectral imaging (30.0%), smartphone-based imaging (19.7%), and calibration and robustness strategies (12.7%). Machine learning methods, particularly convolutional neural networks, support vector regression, XGBoost, and Vision Transformers, increasingly complemented physics-based optical models. Representative studies reported MAE as low as 0.25 g/dL, RMSE of 0.81–1.88 g/dL, correlation coefficients up to r = 0.987, and diagnostic accuracy exceeding 99% under controlled conditions. However, measurement accuracy remained strongly affected by skin pigmentation, motion artifacts, illumination variability, anatomical measurement site, and calibration transferability.
Conclusions: Spectroscopy integrated with artificial intelligence demonstrates strong potential for non-invasive Hb assessment. Nevertheless, standardized validation protocols, multicenter clinical studies, and improved robustness across diverse populations are essential before widespread clinical adoption.
