Integrating deep learning and near-infrared spectroscopy for quality control of traditional Chinese medicine extracts

Abstract

Extraction is a key fundamental step in producing most traditional Chinese medicine (TCM) products. Influenced by factors such as herbs and production techniques, the quality of TCM extracts may vary significantly, leading to high variability between batches and unstable quality of the products. Currently, high-performance liquid chromatography (HPLC) and other analytical methods have been employed for the quality control of TCM extracts, but these methods are time-consuming, environmentally unfriendly, and complicated in sample pre-treatment, which cannot meet the needs of actual production. Therefore, a rapid and effective method for the quality analysis and control of TCM extracts is urgently needed. In this study, the traditional Chinese medicine Radix Gentianae Macrophyllae (RGM) was taken as the research object, and a quantitative analysis model for the active ingredients loganic acid (LA) and gentiopicroside (GPS) was constructed based on near-infrared (NIR) spectroscopy, combining with convolutional neural network (CNN) and gated recurrent unit (GRU). Moreover, the multi-task mean square error (MTMSE) loss function proposed in this paper was used to train the model to ensure the prediction accuracy for different tasks. Meanwhile, this model integrated Bayesian optimization to automatically find the optimal hyperparameters without manual involvement. The results showed that the CNN-GRU model had higher prediction accuracy compared with CNN, BP, and PLS models. In addition, the feasibility of the CNN-GRU model was comprehensively analyzed. Finally, the feature extraction process of the CNN-GRU model was visualized to enhance the interpretability. This research creatively integrated deep learning with near-infrared spectroscopy to achieve rapid and accurate analysis of the active ingredient content of TCM extracts, providing new ideas and methods for quality control of TCM preparations.