肝癌不論在台灣或者世界的死亡率長期名列前茅,而 C 型肝炎為造成肝癌的一個主要因素之一,所以在病患罹患 C 型肝癌後,藉由病患治療前後的基本資料,以及定期追蹤病患的身體狀態,來建立預測病患在未來罹患肝癌的模型,了解該病患是否為日後容易得到肝癌的高風險族群對於醫生及病患都是很重要的。 本文利用台大醫院於 2004 到 2022 搜集的病患資料,總共搜集共 1851 位病患資料,這些病患皆曾罹患 C 型肝癌,並且在之後每隔大約半年進行回診直到該病患罹患肝癌或者停止回診,其中 143 位病患最終罹患肝癌。本文比較多種深度神經網路的模型,加上處理不平衡資料集的方法,建立一個有效預測評估病患未來是否罹患肝癌的模型,並對模型結果做出解釋,提供模型當中重要的特徵指標,希望能對於日後評估 C 型肝癌病患罹患肝癌的風險有所幫助。
The morality rate of liver cancer has ranked high in Taiwan and worlds. Also, hepatitis C is one of the leading cause of liver cancer. Thus, understanding the risk in patients with hepatitis C is important to take care of those patients who has potential to develop HCC in the future. This paper uses the dataset collected in National Taiwan University Hospital (NTUH) from 2004 to 2022. The dataset contains total 1851 patients with hepatitis C virus before. These patients have been cured and continue to be followed-up. Then, they have taken an examination around every 6 months from before treatment until they stopped returning to the hospital or they have developed HCC. 143 of 1851 patients have developed HCC. We propose a solution for predicting hepatocellular carcinoma in patients with chronic hepatitis C using deep learning technique and imbalanced dataset methods. Also, we provide some important features from the result of the model. We hope that this can help the further research on predicting whether the patients with hepatitis C will develop HCC in the future.