Independent Component Analysis-Applications In Multispectral Brain Magnetic Resonance Image Analysis

近來獨立成分分析（ICA）演算法已應用於磁振造影分析。然而，過去文獻對兩個主要問題並沒有討論。一是，臨床磁振造影像通常只有T1WI，T2WI及PDWI三組影像，並不足以讓獨立成份分析分離出不同成份的腦部組織，在此論文我們稱為完備獨立成分分析(over-complete ICA，OC-ICA)。另一問題是利用隨機方式選取初始投影向量作獨立成份分析，如此將造成每次執行獨立成份分析法結果會不一致的情況發生。 為了解決這兩個問題，本論文首先提出利用以空間為基礎之分類方法，如費雪線性區別法及支援向量機，來解決獨立成份分析面臨磁振造影張數不足分析腦內組織獨立成份之問題。根據我們實驗結果顯示，獨立成份分析法結合空間為基礎之分類方法的確解決影像此問題，另外也能解決獨立成份分析結果不一致之情況。雖然仍是隨機方式呈現分類結果，但是腦部三種主要成份，灰質、白質及腦脊髓液都能正確分類出來。 第二種提出解決的方法是擴維度（BEP）技術，其原理是透過非線性方式產生新的磁振造影像，讓影像張數足以利用ICA分析出腦部不同成份的組織。另外，為了解決隨機選取初始投影向量所衍生的問題，我們提出順序獨立成份分析（Prioritized ICA，PICA)讓獨立成份分析法每次產生獨立成份（ICs）都會一致。最後利用BEP與PICA的結合進行磁振造影像分析，結果顯示經過PICA的獨立成份分析法會比傳統直接利用ICA方法來得好。 在整個實驗分類的過程中，本論文發現腦殼會影響腦內組織分類的結果。因此，本論文以去除腦殼探討對ICA及SVM的影響，其結果發現先將MR影像進行ICA分析及SVM分類，再去除腦殼後計算其量化結果為最佳。Independent component analysis (ICA) has found great promise in magnetic resonance (MR) image analysis where signal sources in MR images can be separated by the ICA via its produced Independent Components (ICs). Unfortunately, in order to apply ICA effectively for MR image analysis two key issues must be addressed but have been overlooked in the past. One is the lack of MR images to be used to unmix signal sources of interest, referred to as over-complete ICA (OC-ICA). Another is the use of random initial projection vectors by ICA which causes inconsistent results. This dissertation explores ICA applications in brain MR image classification while addressing the two issues. Since the ICA is a source separation technique and is not developed for classification, this dissertation first develops two ICA-based classification techniques which implements the over-complete ICA in conjunction with spatial domain-based classification, such as Fisher's linear discriminant analysis (FLDA) and Support vector machine (SVM), so as to achieve better classification in each of ICA-demixed ICs. Surprisingly, experimental results show that with the help of classification, the OC-ICA performs significantly better in terms of classification of three major brain tissue substances, WM, GM and CSF. Despite that the three-class classification may appear in different orders resulting from a random order that ICs are generated, such a random appearing order has very little effect on classification results. Secondly, in order to address the issue of insufficient MR band images, a band expansion process (BEP) is proposed and developed to generate an additional new set of images from the original MR images via nonlinear functions. These newly generated images are then combined with the original MR images to provide sufficient MR images for ICA analysis. Thirdly, to resolve the issue of randomly initial projection vectors used by the ICA, a prioritized ICA (PICA) is designed to rank the ICA-generated independent components (ICs) so that MR brain tissue substances can be unmixed and separated by different ICs in a prioritized order. The BEP and PICA are combined to further develop a new ICA-based approach, referred to as PICA-BEP to perform MR image analysis. The experimental results show that PICA-ICA improves the classification performance of traditional ICA approaches and spatial domain-based analysis techniques such as C-means. It has been shown that the brain skull interference which can significantly hinder MR image analysis. So, for the ICA to be effective it needs an additional IC to accommodate the brain skull. Unfortunately, this requirement will further deteriorate the ICA performance due to the lack of MR band images. In order to resolve this dilemma, this dissertation concludes with developing a skull stripping process as a preprocessing so that the brain skull effect can be minimized to enhance the ability of ICA in source separation.Contents 誌謝…………………………………………………………………………… i 摘要…………………………………………………………………………… ii Abstract……………………………………………………………………… iv Contents……………………………………………………………………… vi List of Figures……………………………………………………………… ix List of Tables……………………………………………………………… xii Chapter 1. INTRODUCTION…………………………………………………… 1 1.1. Issues of ICA in MR Image Analysis……………………………… 1 1.2. Organization of Dissertation……………………………………… 4 Chapter 2. BACKGROUND……………………………………………………… 5 2.1. Magnetic Resonance Images………………………………………… 5 2.1.1 Magnetism……………………………………………………………… 5 2.1.2 Resonance……………………………………………………………… 6 2.1.3 Imaging………………………………………………………………… 8 2.1.4 Facilities of MRI…………………………………………………… 8 2.1.5 Pulse sequence………………………….…………………………… 10 2.1.6 Affect parameters of MR image contrast……………………… 11 2.1.7 MR spatial images…………………………………………………… 12 2.2. Independent Component Analysis (ICA)…………………………… 15 2.2.1 Original concepts of ICA………………………………………… 15 2.2.2 FastICA Algorithm…………………………………………………… 19 Chapter 3. OC-ICA WITH CLASSIFICATION………………………………… 24 3.1. Introduction…………………………………………………………… 24 3.1.1 OC-ICA and UC-ICA…………………………………………………… 25 3.1.2 The issue of random initial projection in ICA……………… 27 3.2. Classification to be Used with ICA……………………………… 29 3.2.1. Fisher’s Linear Discriminant Analysis (FLDA)…………… 29 3.2.2. Support Vector Machine (SVM)…………………………………… 30 3.3 Experiments……………………………………………………………… 34 3.3.1 Synthetic Brian Image Experiments……………………………… 34 3.3.2 Quantitative Analysis……………………………………………… 42 3.3.3 Real MR Brian Image Experiments………………………………… 45 3.4 Conclusion……………………………………………………………… 53 Chapter 4. OC-ICA WITH BAND EXPANSION………………………………… 54 4.1. Introduction…………………………………………………………… 54 4.2. Band Expansion Process (BEP)……………………………………… 55 4.3. Prioritized ICA with BEP (PICA-BEP)…………………………… 57 4.3.1 Eigenvector-Prioritized ICA (Eigen-PICA)…………………… 58 4.3.2 High-Order Statistics-Based PICA (HOS-PICA)………………… 59 4.3.3 ATGP-Prioritized PCA (ATGP-PICA)……………………………… 59 4.4. Experiments…………………………………………………………… 61 4.4.1 Synthetic Brian Image Experiments……………………………… 62 4.4.2 Quantitative Analysis……………………………………………… 67 4.4.3 Real MR Brian Image Experiments………………………………… 72 4.5 Conclusion……………………………………………………………… 76 Chapter 5. OC-ICA WITH SKULL STRIPPING AND CLASSIFICATION……… 77 5.1. Introduction…………………………………………………………… 77 5.2. Skull Stripping……………………………………………………… 78 5.3. Noise Issues…………………………………………………………… 79 5.4. Experiments…………………………………………………………… 80 5.5 Conclusion……………………………………………………………… 93 Chapter 6. CONCLUSION……………………………………………………… 94 6.1. Summary………………………………………………………………… 94 6.2. Future Work and Study…………………………………………………96 Reference……………………………………………………………………… 97 Publications..…………………………………………………………………10