體外人體血液中糖化血色素水平之光譜學研究

Abstract

糖尿病的快速增長已成爲全球醫療保健系統的重大挑戰。糖化血色素（HbA1c）作為一種廣泛使用的血糖控制指標，其準確與便捷的量測對於糖尿病的管理至關重要。在光學量測方法中，波長選擇直接影響估測結果的準確性與穩定性。然而，至今尚未有研究系統性分析與比較糖化與非糖化血紅素之光譜特徵和差異，亦未提供具實證依據的波長選擇建議，以支持HbA1c光學量測方法之設計與優化。 本研究旨在探討糖化血色素與非糖化血色素（HbA0/Non-HbA1c）之吸收光譜特性與差異，並初步探討波長選擇對雙波長光學估測HbA1c%效能的影響。研究共分三階段進行：首先，分別對研究級純物質樣本與體外人體血液樣本進行350–650 nm波段之吸收光譜量測與特徵分析，系統比較HbA1c與HbA0之光譜差異。其次，提出混合吸收光譜解混策略（Hybrid Absorption Spectral Unmixing Strategy），結合線性光譜解混法（Linear Spectral Unmixing, LSU）與非負矩陣分解法（Non-negative Matrix Factorization, NMF），從血液樣本中解離出具有代表性的HbA1c與Non-HbA1c成分光譜。最後，根據光譜差異特徵，建立雙波長HbA1c%估測模型，並系統對比不同波長組合對模型估測效能之影響。 研究結果顯示，HbA1c與Non-HbA1c於Soret band peak（415.3 nm）、β-band peak（541.7 nm）與α-band peak（576.6 nm）具穩定且具辨識性的吸收差異（吸收係數Non-HbA1c > HbA1c），並於差異吸收峰601.7 nm處呈現相反趨勢（Non-HbA1c < HbA1c），這些差異吸收特徵為波長選擇提供了關鍵依據。驗證結果顯示，波長選擇對雙波長模型估測效能具有關鍵影響，合理搭配可提升準確性，惟實際應用仍待進一步驗證與優化。 本研究首次系統性比較糖化血色素與非糖化血色素之吸收光譜特性，並從體外人體血液樣本中成功解離出具生理代表性的HbA1c與Non-HbA1c吸收光譜。透過光譜差異分析與雙波長建模驗證波長選擇對估測準確性的影響，為糖化血色素光學量測系統之發展提供重要參考依據。

The rapid rise in diabetes has become a major challenge for global healthcare systems. Glycated hemoglobin (HbA1c), as a widely used biomarker for long-term blood glucose monitoring, requires accurate and convenient measurement for effective diabetes management. In optical measurement approaches, the selection of wavelengths directly affects the accuracy and robustness of HbA1c estimation. However, to date, no study has systematically analyzed and compared the spectral characteristics and differences between glycated and non-glycated hemoglobin, nor has empirical evidence been provided to guide wavelength selection for the design and optimization of HbA1c optical measurement systems. This study aims to investigate the absorption spectral characteristics and differences between glycated hemoglobin (HbA1c) and non-glycated hemoglobin (HbA0/Non-HbA1c), and to preliminarily explore the impact of wavelength selection on the performance of dual-wavelength optical estimation of HbA1c percentage (HbA1c%). The study was conducted in three stages. First, absorption spectra within the 350–650 nm range were measured and analyzed for both research-grade purified samples and ex vivo human blood samples to systematically compare the spectral differences between HbA1c and HbA0. Second, a Hybrid Absorption Spectral Unmixing Strategy was proposed by integrating Linear Spectral Unmixing (LSU) and Non-negative Matrix Factorization (NMF), enabling the extraction of representative HbA1c and Non-HbA1c component spectra from blood samples. Finally, based on the identified spectral differences, dual-wavelength HbA1c% estimation models were developed, and the effects of different wavelength combinations on model performance were systematically evaluated. The experimental results demonstrated that HbA1c and Non-HbA1c exhibited consistent and distinguishable absorption differences at the Soret band peak (415.3 nm), β-band peak (541.7 nm), and α-band peak (576.6 nm), with higher absorption coefficients observed for Non-HbA1c compared to HbA1c. Conversely, an opposite trend was identified at 601.7 nm, where HbA1c showed stronger absorption. These differential spectral features provide critical guidance for wavelength selection in optical measurements. Validation further confirmed that wavelength selection plays a pivotal role in the performance of dual-wavelength estimation model, and that appropriate pairing can enhance accuracy, although further studies are needed to verify and optimize its practical application. This study is the first to systematically compare the absorption spectral characteristics of glycated and non-glycated hemoglobin, and to successfully extract physiologically representative HbA1c and Non-HbA1c spectra from ex vivo human blood samples. Through spectral difference analysis and dual-wavelength modeling, the impact of wavelength selection on estimation accuracy was verified, providing important reference value for the development and optimization of HbA1c optical measurement systems.