Enhanced Technique Improves Precision of Blood Measurement Using Near-Infrared Spectroscopy

A multidisciplinary research team from the Anhui Institute of Optics and Fine Mechanics (AIOFM) and the Hefei Institutes of Physical Science of the Chinese Academy of Sciences (CAS), in collaboration with Hefei Cancer Hospital of CAS, has pioneered a novel approach to accurately quantify hemoglobin levels in blood samples through near-infrared spectroscopy (NIRS). This innovative method aims to address longstanding challenges in blood analysis, mainly the interference caused by water absorption, sample scattering, and instrumental noise, which have traditionally compromised the accuracy of NIRS-based measurements.

Their study, published in Biomedical Signal Processing and Control, demonstrates how this integrated technique can significantly enhance the reliability of blood component analysis. NIRS is renowned for its convenience and non-invasiveness, yet its efficacy in whole-blood testing has been limited by the weak spectral signals of hemoglobin masked by stronger water absorption features. To surmount this obstacle, the research team developed a comprehensive processing strategy combining Wavelet Packet–Fuzzy Shrinkage (WPT-FS) for denoising signals, with the Whale Optimization Algorithm (WOA) to pinpoint the most critical spectral features associated with hemoglobin.

The process involves first applying an adaptive fuzzy shrinkage threshold to eliminate noise across multiple spectral scales, preserving key spectral information. Subsequently, the WOA explores the high-dimensional spectral data to identify optimal feature nodes that capture hemoglobin-specific signals. These features are then used in Partial Least Squares (PLS) modeling to perform accurate quantification.

The approach was validated using 106 whole-blood samples. Results showed a marked reduction in background noise, producing a root mean square error of prediction at 2.0409 and a determination coefficient (R0) of 0.9746. Additional experiments with varied datasets confirmed the robustness of the model, achieving an average prediction error of approximately 1.7%.

This advancement signifies a promising step toward reliable, non-invasive blood diagnostics and monitoring, with potential applications in clinical diagnostics and health management. The research highlights a powerful integration of signal processing and machine intelligence techniques to overcome longstanding limitations of spectral analysis in blood testing.

For further details, refer to the full study: link. Source: https://medicalxpress.com/news/2025-09-method-boosts-accuracy-blood-infrared.html