大腸癌是臺灣常見的癌症死亡原因之一，其多經由息肉演變而來。目前大多以大腸鏡檢查才能夠發現息肉，但大腸內部存在著與息肉型態相似的結構，且傳統上醫學影像需要由有經驗的醫師以人為方式判讀，可能錯誤地解釋其影像或漏診。因此本研究導入自動化檢測，同步以自動判讀方式輔助醫師，獲得可靠且一致性的診斷結果，提高影像診斷與醫療品質。本研究資料蒐集臺灣中部個案醫院大腸鏡影像，首先透過主軸轉換進行影像強化前處理，再以灰階共生矩陣進行特徵萃取九個特徵值，經由統計T檢定，選取六個顯著特徵值（角二次矩、熵、最大機率、對比度、逆差動量、相關性），接續以支撐向量機進行大腸鏡影像資料分類。實驗結果顯示，影像經由主軸轉換與統計檢定選擇後之六個特徵，其分類績效最高，測試組資料之Az值為0.900；績效次優為影像經由主軸轉換與未經選擇之九個特徵，測試組資料之Az值為0.854；分類績效第三為影像未經主軸轉換與統計檢定之六個特徵，測試組資料之Az值為0.800；分類績效最末為影像未經主軸轉換與九個特徵，測試組資料之Az值為0.798。本研究提出可行之主軸轉換、特徵萃取及支撐向量機分類，開發一套大腸息肉自動化檢測系統之演算架構，於醫師進行人工檢測時，同步處理影像及輸出結果，輔助醫師進行判讀，以提升醫療品質。

Colorectal cancer is one of the common causes of cancer death in Taiwan, and most cases of colorectal cancer evolve from polyps. At present, polyps are detected by colonoscopy in most cases. However, there are structures of the similar type to polyp inside the large intestine. Traditionally, medical imaging is interpreted by an experienced physician, which may lead to misinterpretation or omission of the polyps in the acquired images. In this paper, the automatic detection mechanism is developed to assist the physician to improve the quality of diagnosis by providing on-line tools for computer aided diagnosis. The research data were collected from the colorectal images of the case hospital in central Taiwan. The acquired images are enhanced via principle component transformation, and then the grey scale co-occurrence matrix is used to extract features. Six significant features (angular second moment, entropy, maximum probability, contrast, deficit momentum, correlation) were selected by conducting the independent sample t-testing of the nine features. The colorectal classification was conducted by the support vector machines. Experimental results show that the images processed by the principle component transformation and the feature selection perform best. The Az value of the test data set is 0.900. The images processed the principle component transformation and without the feature selection rank second. The test data set's Az value is 0.854. The images without processed by the principle component transformation but processed by the feature selection rank third. The Az value of the test data set is 0.800. The images without processed by neither the principle component transformation nor the feature selection rank lowest. The Az value of the test data set is 0.798. An automatic detection system of colorectal polyps by integrating principle component transformation, feature extraction and support vector machine is developed in this paper to assist the physician in polyps diagnosis to improve the quality of medical care.