以多段三次貝式曲管為基礎的表面模型特徵半自動擷取技術

A Semi-automatic Multi-segment Cubic Bezier Tube Based Feature Extraction Method for Surface Models

林至倫(Chih-Lun Lin)

指導教授 ： 陳永昌

國立清華大學/電機資訊學院/電機工程學系/碩士(2009年)

摘要

腦部是負責掌管人體正常運作的生命中樞。截至目前為止，仍有許多與腦部相關的未解疾病，例如帕金森氏症、阿茲海默症等。然而，人腦結構複雜，是由數十億個神經細胞相互連結，形成一個龐大的神經網路，因此解讀腦神經網路，將會是一個嚴峻的挑戰。在腦科學的研究領域中，果蠅扮演了重要的角色。果蠅除了具有學習與記憶能力，其腦細胞較少(約有十三萬個)且結構較人腦單純，基因結構與人類十分相似，相當適合作為腦科學的研究對象。
隨著共軛焦顯微鏡與染色技術的成熟，我們能夠針對果蠅腦進行精度小於0.1μm的掃瞄取像。為了解讀由不同果蠅腦掃瞄所取得的神經組織，預先建立一個果蠅的標準腦殼來協助腦神經的定位是必須的。然而將單一果蠅腦殼作為標準腦殼並不客觀，最直觀的解決方法是將所有經過三維重建的果蠅腦殼模型做平均的動作，來求得客觀的標準腦殼，最後再將同樣經過三維重建及平均的標準腦神經置入此標準腦殼，便可協助研究人員瞭解果蠅腦神經網路的結構。
由於不同的果蠅間擁有程度不一的變異，故由不同果蠅取像所得到的腦殼模型亦帶有程度不一的變異。可以預見的是此一變異將會造成平均腦殼在某些部位會與單一個體有著不一致的狀態。例如，當不同腦殼間的某些部位帶有窄長且開口方向不一致的凹口時，在經過平均的處理後，得到的平均腦殼會在這些部位形成一個封閉的狀態，明顯與個體有著結構上的差異，這是我們所不樂見的。
本論文提出了一個完整的系統架構解決此問題，在進行腦殼平均動作前，先利用貝式曲線將單一個體中的特徵(如窄長的開口處)做精確的描述，再利用這些特徵曲線作為不同腦模型間區域性形變的參考資訊，進一步對所有將參與平均動作的腦殼做形變的動作。從我們的實驗可以證明，經由調整後所得到的平均模型是更精確的。本論文將著重在如何利用貝式曲線來描述表面模型的特徵，同時利用半自動化技術擷取所有將參與平均動作模型的特徵曲線。實驗結果證明，我們所提出的半自動特徵擷取技術，能夠精確地描述大部分的果蠅腦特徵。

關鍵字

貝式曲線 ； 特徵擷取 ； 表面模型

並列摘要

In basic research of life science, a fruit fly, Drosophila melanogaster, with the abilities of learning and memory is chosen for research to facilitate the understanding of structures and functions of the brain neural network. In the neurobiology of Drosophila, a standard brain (atlas) is necessary. The 3-D surface model of the Drosophila brain is constructed from confocal microscopy scans of individual Drosophila brains. Combining all these individual brain models and applying the model averaging procedure, we can finally generate the 3-D Drosophila brain atlas. However, before applying model averaging procedure, an issue still needs to be considered, i.e., the individual variations of each brain surface model. For example, in our brain surface models, there are some narrow and concave-shaped structures with different orientations in each individual brain model. This would cause structure errors during the model averaging procedure. In our experiments, we could see that there are some close-up structures in the averaged brain atlas, different from the original concave-shaped structures of individual brain models. To solve this problem, we proposed a semi-automatic Bezier curve based feature extraction algorithm for surface models. By using our algorithm, we can extract the features of each individual brain surface model semi-automatically. Furthermore, by applying a general averaging procedure to these feature curves, we can use the averaged feature curves as the reference for local surface warping. Experimental results show that most of the desired features of our brain models could be extracted well, and perform our algorithm before model averaging procedure leads to a more accurate brain atlas. This thesis will focus on the semi-automatic Bezier curve based feature extraction algorithm for surface models.

並列關鍵字

Bezier Curve ； Feature Extraction ； Surface Model

參考文獻

[2] Chao-Yu Chen, “A Framework for Averaging and Updating Surface-Model on Drosophila Brain from Confocal Microscopy Images,” M.S. thesis, National Tsing Hua University 2008.

[3] Nikhil Gagvani, “Parameter-Controlled Volume Thinning,” Graphical Models and Image Processing 61, pp. 149-164, 1999.

[4] Y. Zhou, A. Kaufman and A. Toga, “Three-dimensional Skeleton and Centerline Generation Based on an Approximate Minimum Distance Field,” The Visual Computer 14, pp. 303-314, 1998.

[6] D. Reniers and Jarke J. van Wijk, “Computing Multiscale Curve and Surface Skeletons of Genus 0 Shapes Using a Global Importance Measure,” IEEE Transactions on Visualization and Computer Graphics, VOL. 14, NO. 2, March/April, 2008.

[1] Ying-Cheng Chen, “Framework for Creation the Drosophila Standard Brain from Confocal Microscopy Images,” Ph.D. thesis, National Tsing Hua University, 2007.