使用改良的電荷流體模型實現腦部核磁共振影像的大腦擷取

Abstract

在這篇論文中，我們利用電荷流體模型(Charged Fluid Model，CFM)來實現腦部核磁共振(Magnetic Resonance，MR)影像的分割，並針對電荷流體模型的演算法提出兩個新的權重參數，藉此改進原電荷流體模型演算法在目標物件輪廓模糊時的分割缺點。從概念上來說，電荷流體模型是一個模擬電荷流體行為的封閉曲線，它就像是液體一樣會流過或繞過各種不同的障礙，而在程序上我們將這個概念分成兩個步驟來進行。首先，我們將電荷分佈在特定的傳播界面(Propagating Interface)裡，並將電荷限制在此界面內達到靜電平衡。接著，第二步驟是根據影像強度的影響，將限制電荷分佈的傳播界面做適當的變形。一直重複這兩個步驟即可將曲線停留在目標物件的輪廓邊緣。我們使用此模型方法進行腦部核磁共振影像的分割，並使用數種影像資料庫進行實驗。實驗結果顯示本研究所提出的新權重參數，可以有效地改善原電荷流體模型演算法的分割效果。改進後的方法在各種模擬的雜訊狀況下皆具有相當不錯的結果。在臨床真實核磁共振影像的分割結果，亦有相當高的精準度。

In this thesis, we modify the Charged Fluid Model (CFM) to perform the segmentation of brain magnetic resonance (MR) images. We propose two new stopping forces for the CFM algorithm. Conceptually, the CFM is a simulation of charged fluid, which is like a liquid flowing through and around different obstacles. We divide the process into two steps. First, the CFM flows within the propagating interface until a specified electrostatic equilibrium is achieved. The second step is to evolve the propagating interface based on several image features. Those two procedures are repeated until the propagating front resides on the boundary of objects being segmented. We used this new model for brain MR image segmentation and conducted experiments using a large number of image volumes. The results showed that the new stopping forces can effectively improve the CFM algorithm to segment noisy images as well as real brain MR images.