應用於不同果蠅腦部體積模型三維對位之全自動特徵對應點選取演算法

Automatic Feature Point Detection Between Different 3D Volume Models of Drosophila Brain For 3D Volumetric Registration

向哲緯(Che-Wei Hsiang)

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(2012年)

https://doi.org/10.6843/NTHU.2012.00603

摘要

為了要解開學習和記憶的過程，科學家致力於研究人腦神經網路的互相連結關係，但是人腦的結構複雜龐大，研究非常困難。而果蠅的腦部被發現具有和人類類似的記憶與學習能力。果蠅腦的獨立結構和人腦雖然有很多相異處，但是基因的控制卻和人類相當類似，果蠅大腦結構比人腦簡單許多，並且果蠅具有能夠快速繁殖的優勢，因此科學家將研究轉為果蠅腦的研究。
在研究果蠅腦部機制的過程中，常常需要比較不同果蠅個體間腦部結構或神經網路的差異。在觀察不同果蠅腦部實驗時，會先將欲比較的腦部資料做三維的對位處理。因此在我們的實驗中，希望至少能夠把重要的器官或特徵經過對位處理，使其能落在同一座標位置上。一般三維腦部對位演算法大多是考慮特徵點的資訊，進而對整個三維腦部資料做變形，在我們的實驗中也使用相同的方法，因此我們需要在三維空間中找到這些重要的特徵點。基於人類的視覺無法透視，所以我們很難在三維空間中透過人工的方式取得內部的特徵對應，因此我們希望能夠透過一種演算法來幫助我們自動取得腦內部的特徵對應來達成對位處理。所以在本篇論文中，我們提供了一個自動化提取特徵對應點且適用於果蠅腦部體積模型的對位方法，可以提供良好的對應特徵點並解決可能人工無法精準對應的問題。在我們的演算法中，首先我們必須先找到特徵點並且對應它們，然後我們利用薄板樣條演算法來達成兩個不同腦的對位。處理完畢以後會得到兩個形狀近似且特徵點對齊之果蠅腦部體積模型。

關鍵字

全自動特徵對應點選取

並列摘要

Understanding how people learn and memorize is one of the goal for brain research. In order to simplify research, the main structures and function of Drosophila brain are being intensively studied, because it has been discovered that several brain controlling genes are very similar to human’s.
In order to study the structures and function of Drosophila brain, a volumetric registration process is required to match two volume data of brains. Before volumetric registration is applied, we need to specify features to be aligned during the registration, and then deform the source brain volume into the target one according to these pre-specified features. Therefore, we need to find these feature pairs in 3D volume data of both brains. In the 3D space, it’s difficult to find feature points within the Drosophila brain visually, because we can’t see through the brain volume or other important organs inside. So, we only find feature pairs in each slice respectively.
In this thesis, we develop an automatic feature point detection for Drosophila brains to solve the problem that it’s hard to find corresponding landmarks manually. In our algorithm, first we find feature points and match them globally and locally. Then, we register two different brains with thin-plate spline(TPS). The result will be very similar to target brain, containing the internal organs and characteristics.

並列關鍵字

Automatic Feature Point Detection

參考文獻

[2] David G. Lowe, “Distinctive image features from scale-invariant keypoints’’, International Journal of Computer Vision, 60,2(2004),pp.91-110.

[3] Z. Wang, A. C. Bovik, H. R. Sheikh and E. P. Simoncelli, "Image quality assessment: From error visibility to structural similarity," IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600-612, Apr. 2004.

[4] M. Fischler and R. Bolles, “Random sample consensus : A paradigm for model fitting with applications to image analysis and automated cartography,’’ Communication of the ACM, vol.24,pp.381-395,1981.

[5] F.L. Bookstein, “Principle Warps: Thin-plate splines and the decomposition of deformations”, IEEE Transactions on Pattern analysis and Machine Intelligence, vol. 11, no. 6, 1989

[7] K. S. Arun, T.S. Huang, and S.D. Blostein, “Least-Squares Fitting of Two 3-D Point Sets”, IEEE Transactions on Pattern Analysis and Machine Intelegence, vol. PAMI-9, no. 5, 1987