Identification Models for Chinese Herbal Medicines Using Near-Infrared Spectroscopy

中草藥在現代醫學上的用途越來越多而廣，對其中的多種成份也有許多的科學家進行研究。對製藥廠而言藥材的控管和辨識是很重要的一件工作，如何能快速的辨識中草藥的原料是否為正確的物種與品項，是一個很大的挑戰。本論文針對中草藥，以近紅外光技術進行光譜分析，建立鑑別模式為研究主題。以傳統的化學分析方法作藥材的鑑別，其檢測成本與效率都有改善的空間。近紅外光譜(near infrared spectroscopy)檢測技術與現行其他科學檢測方法相比，具有非破壞性、量測快速等優點，因此可以利用其優點發展為中草藥材快速鑑別之檢測工具。本論文主要包括六個研究，第一部份為背景研究：先使用之50種及100種中藥材原料，包含草藥、菌菇類及動物藥材等進行初步進紅外光譜之分析測試。並依樹狀分析法(hierarchical clustering analysis, HCA)及形質分析法將樣品分群；再以244種藥材為樣品，建立HCA、SVM(support vector machines, SVM)及類神經網路(artificial neural network, ANN)三種分析方法之鑑別模式，結果發現可以使用HCA先將樣品分群後再利用類神經網路建立鑑別模式，為可行之中藥材原料辨識方法。 另因草藥為中藥原料藥材之大宗，草藥的鑑別在中藥製藥與用藥過程非常關鍵，由於藥用植物來源非常廣泛，就算不是近源種，其外觀也可能相似，因此第二部份的研究以中草藥製藥廠的18種原料藥材利用傅立葉轉換近紅外光譜建立鑑別模式，對校正組71個樣本，可以達到100%的辨識率；而對於驗證組的34個樣本，以校正組所建立之模式進行預測分析，可準確辨識33個樣本，其辨識正確率達97%，以FT-NIR進行18種中草藥105個樣本之鑑別可達99%之辨識正確率。第三部份再進一步應用近紅外光及類神經網路技術建立中草藥鑑別模式以辨識不同藥材原料，尤其是無法由其外觀以肉眼判斷之藥材；本研究所建立之模式，是以類神經網路分析中草藥的近紅外光吸收光譜，成功建立可有效區分30種藥品的中草藥粉末鑑別模式，以最佳模式鑑別具有600個樣本的校正組，其正確率為99.67%，鑑別含300個樣本的預測組則可達到100%。 而許多中草藥為了儲藏與流通方便，或是進行科學製藥的前處理，常以粉末型態保存。因此對於此階段的原料藥品進行外觀判斷藥品種類是非常困難的。所以對於藥廠應用於言，建立一項快速而準確的檢測方法與模式就成了很重要的目標與工作。論文的第四個研究：研究利用非破壞性的近紅外光光譜檢測製藥加工前的中草藥原料48種各30個樣品，共1440個樣本數，並利用柔性獨立模式分類分析法(soft independent modeling classification analogy, SIMCA)建立生藥粗原料之定性分類模式。由模式分析可剔除摻假的樣本，樣品辨識率可達到98~100%的辨識率，此方法不僅可以使用於製藥工業，對於食品工業而言，其原料辨識也可以使用此檢測模式及方法，以有效確保樣品的鑑別與摻假的檢測。 本文第五個主題則討論傳統中藥材中時常使用之保育類野生動物穿山甲之圈養調查，彙整自1877以來全世界七種穿山甲的飼養記錄，並討論台北市立動物園飼養穿山甲的歷史及配方的改進，了解穿山甲人工飼養的困難。第六部份：探討如何研究利用近紅光譜學及遺傳學來鑑定含有瀕危動物組成之中藥材原料，利用已建立的穿山甲鱗片及含有犀牛角近紅外光譜資料庫，可以成功區分非屬資料庫的樣品。 最後的補充部份為附錄，附錄的內容主要為探討中藥的定量分析、西洋蔘與人蔘的辨識及人蔘產地及品質的研究以及有毒性之中藥材廣防己與粉防已的近紅外光快速辨識方法的建立等。The inclusion of herbal medicines in modern medical treatment is steadily increasing. Many constituents of the herbal medicines are known and analyzed by scientists. The control and monitoring of the herbal medicine materials is a crucial work in the pharmaceutical factory. Rapid recognition of the plant species and varieties for herbal medicine is a big challenge. The identification by wet chemical methods is usually higher in cost and lower in efficiency. Compared with other modern inspection methods, near infrared (NIR) spectroscopy is an alternative, which is non-destructive, rapid, and easy to operate. There are six main research projects in this dissertation. Firstly is the previous backgrand study: the ability of using NIR spectroscopy in differentiating from 50 to 100 different herbal medicine raw materials was demonstrated. The examples included a variety of samples based on plant, fungi and animal derived materials. In addition, to simplify the identification, the author used hierarchical cluster analysis and other pattern recognition techniques, groupings of similar materials (based on NIR spectra, not priory groupings). The identification for the products in each grouping could then be definitively made using pattern recognition techniques that were customized for each distinct group of materials. Hierarchical clustering analysis (HCA), artificial neural network (ANN) and support vector machines (SVM) were applied to 244 herbal medicine raw materials classification problems by NIR. The SVM training resulted in models showing a method for the identification of herbal medicine raw materials. The results indicate when (HCA) distances were computed, 10 PCs were used to cover 95% of information. If threshold 1 is used, library will be divided into 16 clusters. The following different clusters can be used for the local ID methods developments. Clustering total samples into secondary groupings make identification more definitive. When using the feed forward network as a classifier, choose output neurons as many classes in the calibration dataset, each of the output neurons are set to react for only one specific material, if connected to the same hidden layers showed the better results on the same training times calibrations. In the sencond study, 18 raw materials from Herbal Medicines industry were used to examine the FT-NIR performance. Regarding 71 samples in the calibration set, the identification accuracy was 100%. In the validation set of 34 samples, 33 samples were successfully discriminated, and the identification accuracy was 97%. As a result, the identification accuracy of 18 medicinal herbs with 105 samples was 99% using FT-NIR spectroscopy. Furthermore, a robust identification model for herbal medicine was developed by combining near-infrared (NIR) spectroscopy and artificial neural network (ANN) to discriminate raw materials of herbal medicine, which are often similar in appearance and practically impossible to identify by visual inspection alone. The third part research was employed ANN to analyze the absorption spectra of herbal medicines and successfully built an identification model, which is able to identify 30 different herbal medicines. The best identification model can reach a correct identification rate (CIR) of 99.67% when applied to a training set of 600 samples, and 100% CIR when applied to a test set of 300 samples. Moreover, because the storage and conveyance of the raw materials are always in dry powder forms before the materials are used in scientific pharmaceutical procedures, it is difficult to determine specific varieties of the herbal medicine constituents by visual observation of the raw materials at this step. Consequently, the development of a rapid and accurate inspection method and model for pharmaceutical factory applications is greatly needed. The fourth, a variety of herbal medicines according 48 raw materials with every material content 30 samples were measured using non-destructive near-infrared spectroscopy with soft independent modeling classification analogy (SIMCA) to build up the classification model. The adulterated samples could be eliminated by the analysis of the model, and identification rates were demonstrated in the range of 98 to 100%. The method could be applied not only to the pharmaceutical industry but also to the food industry. The food materials could be measured with the inspection model for effective identification and determination of adulteration. The fifth study: because the traditional Chinese medicine (TCM) is the quintessence of Chinese culture with long history. It has, for many years, enriched the quality of people’s life. TCM is mainly obtained from nature. Unfortunately, some of the nature resources are no longer sustainable due to habitat destruction and over-exploitation. Many of the animals and plants that are used in TCM have become endangered. The objects of this part study was to establish a history of feeding and dietary husbandry of pangolin in captivity since 1877 to 2001, and discuss the methods of identify those endangered species which used in TCM by NIR and DNA analysis in the six part. Finally, in addition on appendixs: there were 3 main studies in the appendix：1. Quantitative analysis of herbal materials, 2. Genseng and western ginseng study: species differentiation, geographical origins, quality assessment and 3. Fangij compare with the poisonous herbal material guangfangji by NIR