應用於MRI影像之適應性權重模糊分割演算法

Abstract

磁共振影像由於是安全無放射性且具有高對比成像，所以在現今醫療的診斷上佔有重要的地位，且成為常見的診斷工具之一。然而在醫療影像上常會帶有不均勻的雜訊，造成影像品質不佳，導致醫生在診斷上較為不便，而影像上雜訊也會對醫生的診斷造成影響。影像的分割是透過抑制雜訊的影響並將腦部中各組織分割開來，得到區塊分明且可視性佳的影像，目的是為了在診療上輔助醫生做判斷。本論文提出一種對於雜訊的影響程度低且分割準確度高的模糊分類演算法，利用模糊分類的概念，在分類上使用歸屬度來表示歸屬於分群的程度，可以更準確地描述分類情況。對於不均勻雜訊，則利用適應性的權重設定對於像素資訊做適當的擷取，可以降低雜訊影響造成分類錯誤。而適應性權重的歸屬度更新，是透過鄰近區域之歸屬度為參考，將歸屬度分布做更新，能夠有效排除區域上的特異像素點以及不均勻分布的灰階值，可以得到正確的分群方向，進而得到較佳的分割表現。

Magnetic Resonance Imaging is safe and non-radiation. There is good contrast in MRI. Therefore, it plays a very important role in the medical diagnosis. There are many ununiform noise in medical images. Thus the quality of images are poor. The noise in images interfere the determine of doctor cause doctor is inconvenient. The image segmentation is to segment the tissue of brain by suppress the influence of noise. We gain the image which is clearly and visible to help doctor to diagnose. This paper have proposed an algorithm which is low-interference with noise and accurate in segmentation. In clustering, we use the concept of fuzziness to express the degree which is belong to the cluster center by membership. Because of it, we can describe the situation of clustering accurately. To reduce the effect of ununiform noise, we use adaptive weight to control the importance of pixel for clustering. Thus, the mistake of clustering will be decrease. The membership are updated by referencing the membership of neighbor pixel. In this way, we can remove the effect of noise, and we can get the image with better performance of segmentation.