Accurate Estimation of Methemoglobin and Oxygen Saturation in Skin Tissue Using Diffuse Reflectance Spectroscopy and Artificial Intelligence

• Published in: Journal of biophotonics Vol. 18; no. 5; p. e202400413
• Main Authors: Sahli, Isra, Bachir, Wesam, El‐Daher, Moustafa Sayem
• Format: Journal Article
• Language: English
• Published: Germany Wiley Subscription Services, Inc 01.05.2025; WILEY‐VCH Verlag GmbH & Co. KGaA
• Subjects: Artificial Intelligence; Artificial neural networks; Broadband; Diffuse reflectance spectroscopy; Hemoglobin; Humans; Machine learning; Methemoglobin; Methemoglobin - metabolism; Neural networks; Neural Networks, Computer; Oxygen; Oxygen - metabolism; Oxygen content; Oxygen Saturation; Reflectance; Saturation; Skin - metabolism; Spectra; Spectroscopy; Spectrum Analysis - methods; artificial neural networks; Monte Carlo simulations; diffuse reflectance spectroscopy; methemoglobin
• ISSN: 1864-063X; 1864-0648; 1864-0648
• DOI: 10.1002/jbio.202400413

In this paper, we present a noninvasive method for the accurate estimation of methemoglobin concentration. The proposed technique incorporates a novel machine learning model using the artificial neural network to detect methemoglobin and oxygen saturation from the diffuse reflectance spectra of skin tissue. Sixty‐six spectra were simulated using a four‐layer tissue model with varying oxygen saturation and methemoglobin concentration. A multifiber probe‐based DRS setup in the visible and near‐infrared wavelength range was used. The best accuracy, with a mean absolute error (MAE) of 0.0392% for the concentration of methemoglobin and 0.0273% for the percentage of oxygen saturation on the created data set, was achieved. Our method was also experimentally verified using DRS spectra collected from human subjects. Consequently, the findings demonstrate the ability of broadband DRS to noninvasively differentiate subtle changes in methemoglobin and hemoglobin levels despite their overlapping spectral features.