Please use this identifier to cite or link to this item: http://rportal.lib.ntnu.edu.tw:80/handle/20.500.12235/111984
Title: 開發自動化讀取與分類文獻資料以探討植物性雌激素對婦女癌症的影響
Developing the Automatical Method to Review and Classify Literatures on the Effect of Phytoestrogens on Women Cancers
Authors: 謝佳倩
Hsieh, Chia-Chien
林冠均
Lin, Guan-Jun
Keywords: 婦女性癌症
植物性雌激素
自然語言處理
分類
資料探勘
women cancer
phytoestrogen
Natural Language Processing
classification
text mining
Issue Date: 2020
Abstract: 乳癌、子宮內膜癌及卵巢癌為婦女性癌症中高發生率的惡性腫瘤,流行病學資料顯示其發生率有逐漸增長的趨勢。婦女性癌症和雌激素的調節與作用息息相關,而經由飲食攝取所含的天然雌激素也可能為影響因子之一,植物性雌激素結構與人體製造的雌二醇相似,可能執行類似或拮抗的功效進而干擾體內雌激素的作用。近年智慧醫療相關產業發達,有越來越多研究是利用電腦資訊技術進行,以協助醫療能夠更加進步、精確與優化。本研究運用電腦自然語言學習工具建立自動化分類程式,對摘要內容以分句方式配合分類關鍵字與訓練資料來進行分類判斷,以幫助文獻資料的分類與整理,減少人力閱讀花費的時間與精力,利用植物雌激素類型、癌症類型、實驗類型這三個類別變項建立資料庫,共包含937篇文獻摘要分為1683筆資料,再使用資料探勘技術及決策樹模型進行資料的分析與統合,接著針對癌症種類如乳癌、子宮內膜癌與卵巢癌,和植物性雌激素類別包含香豆酚 (coumestrol)、大豆苷元 (daidzein)、雌馬酚 (equol)、金雀素黃酮 (genistein)、木酚素 (lignan) 與白藜蘆醇 (resveratrol) 介入研究,以及實驗類型如細胞、動物與人體實驗,進而探討植物雌激素介入後對婦女癌風險發展的相關性。本研究利用卡方檢驗及資料探勘模型如決策樹模型及關聯規則,來分析植物性雌激素介入對婦女性癌症的發展影響。在不同類別植物性雌激素介入,發現植物性雌激素lignan與resveratrol對於婦女性癌症的影響多為正面,而daidzein 與equol造成的負面影響相對較多。針對癌症種類討論,其中乳癌對於genistein、daidzein或equol的使用可能會有較高風險,而卵巢癌在植物性雌激素介入後多是正向反應,而子宮內膜癌則含有較多負面結果,特別是對於daidzein或equol的使用。在實驗類型中,細胞或動物實驗顯示相對較高的負面反應,而人體實驗則為中性結果較多,特別在乳癌的研究。綜合上述,本研究建議乳癌與子宮內膜癌的患者在植物性雌激素的食物攝取上可能會有些微的風險,特別在daidzein或equol的單品使用。
Breast cancer, endometrial cancer and ovarian cancer are malignant tumors with a high incidence and have high mortality in women. Epidemiological studies show that the incidence of these types of women cancer is increasing nowadays. Breast cancer, endometrial cancer, and ovarian cancer are closely related to the estrogen status. These phytoestrogens have similar structure with the human estradiol, thus, have similar effects and may interfere the action of estrogen in our body. Recently, the smart health industries use computer technologies to develop medical therapy to achieve advanced, accurate and optimized pathway. In this study, a natural language processing system was used to establish an automated classification program to help data classification, reduce the time and effort of the manual reading. We established a database using four classification categories of phytoestrogens, cancers, experiments and research results. The data base was set up contained 937 abstracts including 1683 data. Then we use text mining techniques and decision tree models to analyze the data. We collected three types of women cancer including breast cancer, endometrial cancer, and ovarian cancer, six phytoestrogens, including coumestrol, daidzein, equol, genistein, lignin and resveratrol. The research type was included cell, animal and human studies. We used chi-square and text mining methods, such as decision tree models and association rules, to explore the correlation between phytoestrogens and women cancer development. Regard to the types of phytoestrogens, we found that the phytoestrogens lignin and resveratrol have a higher positive effect on women cancer, while daidzein and equol are more related to negative effects. According to the type of cancer, coumestrol, genistein, daidzein or equol may have more risk on breast cancer, besides there are more positive reactions on ovarian cancer, and more negative results related to endometrial cancer. In the experiment types, cell and animal experiments showed higher negative impacts, while human experiments have more neutral results, especially in breast cancer researches. In conclusion, patients with women cancer should pay more attention to using daidzein and equol. Thus, patients with endometrial or breast cancer may have a slight risk on intake of foods containing coumestrol, daidzein, equol and genistein, especially daidzein and equol.
URI: http://etds.lib.ntnu.edu.tw/cgi-bin/gs32/gsweb.cgi?o=dstdcdr&s=id=%22G060751008S%22.&%22.id.&
http://rportal.lib.ntnu.edu.tw:80/handle/20.500.12235/111984
Other Identifiers: G060751008S
Appears in Collections:學位論文

Files in This Item:
There are no files associated with this item.


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.