Hyperspectral Camera Imaging Tandem With HPLC‐UV‐ELSD Determination and Cell Bioassay for Geographical Discrimination and Quality Consistency Evaluation of Anoectochilus roxburghii
Haixia Xu, Qiuya Zhou, Qiluo Ni, Weiyue Hu, Xiangwei Xu, Yi TaoABSTRACT
Introduction
Anoectochilus roxburghii (AR) is a prized medicinal herb valued for its hepatoprotective effects. Its quality varies depending on geographical origin. The primary bioactive constituents include rutin, quercetin‐7‐ O ‐glucoside, kaempferol‐3‐ O ‐rutinoside, narcissin, quercetin, and kinsenoside. A method that enables simultaneous determination of both the content and bioactivity of the herb is therefore essential for effective quality control.
Objective
To develop a rapid, nondestructive approach for simultaneously predicting the contents of primary bioactive constituents and hepatoprotective effects of AR using hyperspectral camera imaging (HCI) combined with deep learning models.
Method
Hyperspectral images of 100 AR batches were acquired using a portable Vis–NIR HCI system (389.81–1048.18 nm). The contents of six active compounds were quantified via HPLC‐UV and HPLC‐ELSD, while hepatoprotective activity was evaluated using an APAP‐induced L02 cell injury model. Quantitative calibration models were constructed using partial least squares regression (PLSR) and the following three deep learning architectures: liquid neural network (LNN), Mamba state space model, and graph convolutional network (GCN). Their predictive performances were systematically compared. Shewhart control charts were employed to visualize batch‐to‐batch quality variation.
Result
The Mamba model demonstrated superior predictive performance across all seven quality attributes, achieving the highest coefficients of determination ( R p 2 up to 0.9972) and the lowest prediction errors. It significantly outperformed PLSR, LNN, and GCN models. Furthermore, the integration of Shewhart charts enabled effective visualization of quality consistency across batches.
Conclusion
This study presents a novel HCI‐Mamba framework designed for the rapid, nondestructive, and multicomponent quality assessment of AR. The proposed strategy offers a high‐throughput solution for AR quality control while providing a methodological framework applicable to other complex herbal medicines.