應用“獨立成份分析法”改善腦部MR影像組織分割績效之研究

Improving Tissue Classification Effects of Brain MRI Segmentation Based on Independent Component Analysis

李三剛(San-Kan Lee) ； 蔡志文(Jyh Wen Chai) ； 陳啓昌(Clayton Chi-Chang Chen) ； 陳享民(Hsian-Min Chen) ； 王士偉(Shih-Wei Wang) ； 歐陽彥杰(Yen-Chieh Ouyang) ； 楊晴雯(Ching-Wen Yang) ； 張建禕(Chein-I Chang)

《中華放射線醫學雜誌》 34卷2期 (2009/06) Pp. 93-101

摘要

本論文旨在開發一套裝電腦輔助系統，應用於MR影像腦部組織分割，以期正確執行之腦部灰、白質體積衡量，而有助於腦部定量形態學之研究。在標準合成MR 影像的實驗結果發現，使用隸屬於多頻譜影像分析技術之獨立成份分析(Independent Component Analysis, ICA)方法搭配「支援向量器(Support Vector Machine, SVM)」之演算法，能有效分割MR影像腦部組織，再配合「分水嶺演算法(Watershed Algorithm)」去除非腦組織後，在背景雜訊為0%和3%等兩種MR影像中，其灰／白質分割效果量化指標(Tanimoto Index)分別為0.82/0.89和0.73/0.80，結果優於過去文獻其他技術的報告。正確及客觀的腦部容積衡量是臨床醫學和基礎研究

並列摘要

In this paper we develop a computer-aided brain tissues classification system for MR images and hence make possible for more accurate volume measurement on gray/white matter, and cerebrospinal fluid. A correct classification of brain tissues is an important step in quantitative morphological study of brain. From the synthetic brain MR images experiment, it shows that using multispectral image processing technique, independent component analysis (ICA), and coupling with support vector machine (SVM) method can effectively classify brain tissues for brain MR images. In addition, we also demonstrated that the best performance can be achieved by using watershed algorithm as a pre-processing method for striping non-brain tissues. The Tanimoto index of GM/WM in synthetic MR images with noise level 0% and 3% are 0.82/0.89 and 0.73/0.80, separately which shows better performance than those what we have seen in the literatures.