利用計算流體力學模擬肺動脈壓力來分析與肺動脈高壓症狀之間的關係

Abstract

肺動脈高壓是一種嚴重的肺部血管疾病，因為肺動脈構造改變，使得肺動脈內的血液壓力持續增加。患有肺動脈高壓的病人會覺得呼吸困難、感覺胸痛等。在臨床上，一連串的檢查例如心臟超音波和電腦斷層掃描確認病人是否患有肺動脈高壓、以及利用右心導管檢查測量肺動脈血壓高低。雖然這些屬於標準的肺動脈檢查程序，但是病患可能因為沒有特殊症狀而無法順利進行診斷。在本篇研究中，利用電腦斷層掃描影像重建三維肺動脈結構，並利用計算流體力學軟體去模擬肺動脈中血液壓力分佈。本研究目的建立於利用CFD模擬軟體建立出的肺動脈壓力分佈，藉由從中獲取的壓力特徵值來增加肺動脈高壓預測模組的準確率。 程式首先，利用半自動切割法將電腦斷層掃描影像中切割出肺動脈範圍建立成一個三維肺動脈模型，再標示三維肺動脈模型的出入口後，接著使用計算流體力學模擬肺動脈血液壓力分佈，最後，擷取壓力特徵來模擬肺動脈壓力分佈以及擷取體積特徵來模擬肺動脈結構並進行分析。實驗結果發現，藉由加入壓力特徵值可提高預測病人是否患有肺高壓準確率，原先使用兩個壓力特徵值的預測準確率可達77.78%，若是加上四個壓力特徵值的組合能夠提高準確率至83.33%。最後使用特徵選擇法選擇較佳的特徵組合將預測準確率最高提高至88.89%。

Pulmonary hypertension (PH) is a serious lung disease nowadays. Because the pulmonary artery (PA) structure is changed, resulting in the increase of PA blood pressure. Patients suffering from PH are difficult in breathing and fell chest pain. In clinic, several examinations, such as echocardiograph and computed tomography (CT) are used to examine the PH patient, and right heart catheterization is used to measure the PA blood pressure for PH diagnosis. Although these examinations are standard medical procedures, patients would not be diagnosed easily without obvious symptoms. In this study, we analyzed the PA pressure distribution on a three-dimensional (3-D) PA model built from a CT scan by using a computational fluid dynamics (CFD) software. The aim of this study is to offer a PH prediction model based on the pressure-based features extracted from the PA pressure distribution by the CFD simulation, and to increase the PH prediction accuracy. First, a semi-automatic segmentation was presented to obtain a 3-D PA model from a CT scan. After setting inlet and outlet planes on the 3-D PA model, the CFD simulation was performed to simulate the blood pressure distribution. Then, the pressure-based and volume-based features were used to extract the characteristics of the simulated PA pressure distribution and the PA structure, respectively. The experimental results showed that the PH prediction accuracy can be improved by using the pressure-based features. Originally, the PH prediction accuracy was 77.78% by using the combination of two volume-based pressures. Then, the accuracy was increase to 83.33% by using the combination of four pressure-based features and two volume-based features. Finally, the feature selection was used to select an optimal feature combination to obtain the highest accuracy of 88.89%.