Rapid detection of multi-indicator components of classical famous formula Zhuru Decoction concentration process based on fusion CNN-LSTM hybrid model with the near-infrared spectrum.

[Display omitted] • Using NIRS combined with CNN-LSTM to monitor the entire process of the classic formula Zhuru Decoction. • CNN-LSTM can accurately and quickly monitor the main pharmacological components of Zhuru Decoction. • The method can provide a reference for the rapid testing of traditional Chinese medicine compound prescriptions. Traditional Chinese Medicine (TCM) is a collective term for the medicines of all ethnic groups in China, including Han Chinese and ethnic minority medicines, reflecting the Chinese people's understanding of life, health and disease, and is a system of medicine with a long historical tradition and unique theories and technical methods. Due to the nonlinearity, large hysteresis, and strong interference in the concentration process used in the production of traditional Chinese medicine, it is challenging to use just one conventional control method to achieve the desired control effect. The majority of traditional quality control methods used in processing the samples will result in the loss of the active ingredient and in the detection of impurities. Monitoring of multiple critical quality attributes (CQAs) is crucial for the overall process of the index ingredient. In this study, based on the above background, fast and nondestructive near-infrared spectroscopy (NIRS) technique incorporating convolutional neural network (CNN) and long-short-term memory network (LSTM) was utilized to construct a quantitative model based on the seven characteristic components, puerarin, 3′-hydroxy puerarin, 3′-methoxy puerarin, puerarin apigenin, daidzin, glycyrrhizin and glycyrrhizic acid, during the concentration process. The results showed that the relative mean error (MAPE) values of the different components ranged from 0.7% to 2.8% and R2 from 0.981 to 0.992 based on the proposed CNN-LSTM model. The results showed that the combined model of near-infrared spectroscopy fused with convolutional neural network (CNN) and long-short-term memory network (LSTM) was feasible for accurate and rapid monitoring of the concentration of the classic famous prescription Zhuru Decoction (ZRD) during the decoction process and that the method could provide a reference value for rapid nondestructive testing of traditional Chinese medicine compound formulas. [ABSTRACT FROM AUTHOR]