據衛生福利部國民健康署民國99年癌症登記年度報告,指出當年因大腸直腸癌惡性腫瘤死亡人數占全部惡性腫瘤死亡人數的11.39%。死亡率排名位居第3位,發生個案數位居臺灣惡性腫瘤中第一位。 臺灣某醫學中心2007-2009年癌症登記資料庫統計,大腸直腸癌第二期病患曾一度攀升佔所有大腸直腸癌各期別百分比最高,其次才是第三、四期。 本研究收集臺灣某醫學中心283位病患92年1月至98年12月的資料,應用決策樹、類神經網路技術及邏輯斯迴歸等機器學習,建構大腸直腸癌第二期存活預測模式。研究發現,資料探勘所建立的模式以類神經網路預測大腸直腸癌第二期最佳,正確率為74.56% (ROC 曲線0.83%),其次為決策樹,正確率為73.14% (ROC 曲線0.72%),最後為邏輯斯迴歸,正確率為71.73% (ROC曲線0.75%)。 依這三種模型可預測大腸直腸癌第二期病患最終結果為存活或無法存活,且提供臨床上大腸直腸癌第二期病患的存活預測,並提供醫師對患者診療及預後評估之建議與參考。
According to the annual cancer registration report by the Health Promotion Administrative, Ministry of Health and Welfare in 2010, the deaths of Colorectal Cancer Malignancies was 11.39% of all malignancies. The ranking of the death rate was the third, and the number of cases was the first among the malignancies in Taiwan. Based on the statistics by a medical center in Taiwan, the cancer registration data base from 2007 to 2009, the patients in the stage II of Colorectal Cancer ranked the highest percentage ; the stage III was the second, and the stage IV was the third. This study collected the data of 283 patients from January 2003 to December 2009 in a medical center in Taiwan and applies the Decision Tree (DT), the Logistic Regression, and the Artificial Neural Networks (ANNs) to construct the survival prediction model of the Colorectal Cancer. The current study found that the constructed model of the Data Mining was the most appropriate by utilizing the ANNs. The accuracy was 74.56% (the ROC curve was 83%). The accuracy of the Decision Tree ranked second which was 73.14% (the ROC curve was 72%). The last one was the Logistic Regression and the accuracy was 71.73% (the ROC curve was 75%). According to these three models, eventually the survival of the patients in the stage II of the Colorectal Cancer can be predicted. These three models provide the survival prediction for the patients in the stage II of the Colorectal Cancer and provide suggestions and references for doctors to conduct treatment and evaluation.