肝癌不但是全世界最常見的癌症之一,而且根據衛生署過去二十多年來的統計,肝癌一直高居國內癌症死亡原因前幾名。而肝臟中的腫瘤可分成惡性和良性。不同的癌化程度其治療方式與疾病預後會有相當大的差異,因此必須利用各種方式來做最正確的鑑別診斷。雖然超音波、X光和核磁共振等的檢測方法可以檢測出肝臟病變,然而肝臟的病理切片卻能提供肝癌最準確的資訊。 然而醫師所提供的腫瘤細胞的分級結果是主觀的、非量化的,因此常會因個人認定不一致的結果,尤其是在一些非典型分化的肝癌細胞更是容易有分歧的意見。這篇論文我們採用支持向量機以及深度學習的方法對腫瘤病理切片的影像進行分類,建立一個客觀的評量方法,可以提供醫師做為診斷的參考得到更準確的診斷。根據我們的實驗結果,我們的方法可以達到97.45%的分類正確性。
Liver cancer is not only one of the most common cancers in the world, but according to the statistics of the Department of Health and Welfare over the past 20-plus years, liver cancer has also been one of the top causes of cancer deaths in Taiwan. The tumors in the liver can be divided into malignant and benign. For different tumor stages, the treatments and prognosis of disease will also be considerably different, so a variety of methods must be adopted to determine the stage of the tumor accurately. While testing methods such as Ultrasound, X-ray and Nuclear Magnetic Resonance (NMR) can diagnose liver lesions, the liver biopsy can provide the most accurate information about the liver cancer. However, the determination of cancer stage provided by the physicians is subjective and non-quantifiable, so such determinations often vary from physician to physician due to different personal opinions - especially in some cases of atypical division of liver cancer cells. This study adopts the Support Vector Machine (SVM) and deep learning method to classify the images of tumor biopsy, thereby establishing an objective assessment method to provide the physicians a diagnostic reference for a more accurate determination of tumor stages. According to our experimental results, the method proposed in this study can achieve a classification accuracy of 97.45%.