EM最佳化結合多閥值階層集合模型於腦部影像分割之應用

Optimized EM with Multi-Threshold Level Set Model approach for MRI Image Segmentation of Brain

蕭詠翰(Yung-Han Hsiao)

指導教授 ： 白炳豐

國立暨南國際大學/管理學院/資訊管理學系/碩士(2011年)

摘要

圖形辨識是機器學習 (Machine Learning) 重要的一個分支，其應用之範圍相當廣泛，也是許多學者亟欲探究的一個學術領域。許多學者採用期望最大值分群演算法（Expectation- Maximization Algorithm，EM）對腦部核磁共振圖(Magnetic Resonance Imaging，MRI) 進行分析，但是如何妥善決定EM演算法中E步驟與M步驟的最佳參數卻很少進行探討，故本研究採用粒子群最佳化演算法（Particle Swarm Optimization，PSO）、禁忌搜尋演算法（Tabu Search Algorithms，Tabu）、免疫演算法（Immune Algorithms，IA）、基因演算法（Genetic Algorithms，GA）等啟發式演算演算尋最佳參數並避免落入區域最佳解 (Local Optimized) 的窘境。在資料的前處理部分，去除不需要的頭蓋骨和脂肪組織後，主要分成三類，包括：白質（White Matter，WM）、灰質（Gray Matter，GM）和脊髓液（Cerebrospinal fluid，CSF）等三個部分，進而採用C. Y. Hsu 等人 (2010) 所提之多閥值階層集合模型（Multi-Threshold Level Set Model）進行影像分割與分析，並比較各種啟發式演算法所得出之正確性與收斂速度。此外，為了驗證本模型的優越性，並進一步比較多種不同影像分割的方法，如Multi-Threshold Level Set Model、Adaptive Threshold Level Set Without Edge Model、Level Set Model等三個影像分割之模型和直接利用分群演算法對MRI影像形做分割的方法，結果皆顯示MTLS具有優越的圖形辨識度。

關鍵字

啟發式演算法 ； 影像分割 ； 階層集合模型 ； EM演算法 ； 圖形辨識

並列摘要

Pattern recognition was one of the main streams of machine learning research and performed a satisfactory job in many research domains. Most parts of researchers utilized EM algorithm to analyze the magnetic resonance image segmentation of brain, however fewer researchers focused on determining inherent parameters in EM algorithm. The suitable parameters setting would have considerable impact on computation cost and classification performance. Thus, the study applied meta-heuristics (particle swan optimization, tabu search, immune algorithm, and genetic algorithm) to optimize the parameters determination. After undergoing data preprocessing, the image would divided into three parts namely, white matter, gray matter and cerebrospinal fluid and sequentially fed into multi-threshold level set model (Hsu et al.,2010) to perform an pattern recognition process. The performance criterion were classification accuracy and converge speed were used to evaluate the efficacy of those four meta-heuristics approach. Moreover, to examine the feasibility of the model we proposed, we further compared other pattern recognition approaches (multi-threshold level set model, adaptive threshold level set without edge model and level set model). According to the experimental result, the presented model possessed an outstanding performance is pattern recognition and quite suitable to cope with the related tasks.

並列關鍵字

Meta-heuristics ； Image Segmentation ； Level Set Model ； EM algorithm ； Pattern recognition

參考文獻

[1] B. Fischl, D. H. Salat, E Busa, M. Albert, M. Dieterich, C. Haselgrove, A. van der Kouwe, R. Killiany, D. Kennedy, S. Klaveness, A. Montillo, N. Makris, B. Rosen and A. M. Dale, “Whole Brain Segmentation：Automated Labeling of Neuroanatomical Structures in the Human Brain”, Neuron, Volume 33, Issue 3, January 2002, Pages 341-355

[2] M. S. Atkins and B. T. Mackiewich, “Fully automatic segmentation of the brain in MRI”, IEEE Transactions on Medical Imaging, Volume 17, Issue 1, February 1998, Pages 98-107

[3] P. D. Sathya, R. Kayalvizhi, “Optimal segmentation of brain MRI based on adaptive bacterial foraging algorithm”, Neurocomputing, March 2011

[4] J. L. Marroquin, B. C. Vemuri, S. Botello, F. Calderon, and A. Fernandez-Bouzas, “An Accurate and Efficient Bayesian Method for Automatic Segmentation of Brain MRI”, IEEE Transaction on Medical Image, Volume. 21, Issue. 8, August 2002, Pages 934-945

[5] R. C. Eberhart and J. Kennedy, “A new optimizer using particle swarm theory”, International Symposium on Micro Machine and Human Science, October 1995, Pages 39-43