腦神經結構重建及發育分析

Abstract

神經的網路結構一直是生物學家感興趣的研究課題。研究者相信神經間的拓樸結 構可以和神經傳導訊息的傳遞有關，並且也和主宰了那些功能有關。透過近代的 基因標定技術以及光學儀器，我們可以清楚的看到單一個帶有螢光表現的神經細 胞，使得我們有機會做更進一步的神經結構分析；不論是單一神經型態或是神經 網路的拓樸。但是經由光學儀器取得的原始影像資料經常是十分龐大的，會帶給 後續的分析很大的困擾，為了解決這個問題，本論文的第一部分將把重點放在如 何萃取出神經的中線。當我們把一個神經的結構從影像轉為點和線的結構後，檔 案的大小將從原來的數百萬位元組降至幾萬位元組，這得以加速後續的分析。我 們同時對提出的方法進行強韌度及準確率測試。結果顯示我們的方法是可靠的。 另外，我們也提出一個基於萃取出的中線結構的應用以證明這些萃取出來的中線 可用於更一進步的結構分析。神經發育是另一個讓許多生物學家及神經科學家投 入精力的課題。透過觀察神經的發育，我們可以理解究竟是甚麼樣的物質及環境 使得神經退化或加速發育。神經的中線結構是科學家們觀察的神經發育的樣本， 但傳統上為了觀察發育的現象只能靠生物學家曠日廢時的以人工標定每個細微處 的變化，這大大的延宕了實驗的進度。本論文的第二部分提出一個監督式的半自 動互動系統(4D SPA)。我們實驗記錄了在4D SPA 的輔助下，使用者需要花多少 時間完成配對標定的工作，同時也採取了兩種在圖形辨識常用的策略來自動標定 配對，並記錄了計算所需的時間。透過比較兩種不同方法，我們確認了相較於傳 統的手動標定，4D SPA 使得標定神經發育這件工作更省時，相比於自動化的方 法，我們的方法不僅省時也更穩健。

Biologists and neurologists have been interested in neural networks for a long time. They believe the network topology is highly related to the neural functions. Modern biotechnology and optical apparatus make it possible to observe a single neuron in micron scale. This gives us the chance to understand the neural circuits and neuronal morphology more. However, the size of the original image data is huge and this largely obstruct the further analysis. Hence, in the first part of this dissertation, we focus on the issue about how to reconstruct the 3D neuronal structures to sets of points and lines. When the data are reduced to points and lines, the data size becomes smaller, from hundreds megabytes to some megabytes. The robustness and accuracy test were performed and the results showed the proposed method is reliable. An application of the reconstructed results was also demonstrated. This showed the further analysis based on the reconstructed results is possible. In conclusion, the proposed method facilitates researchers to extend their study to the higher level of neural structures and the neural networks. The study of neuronal development is also a topic that attracts many biologists and neurologists. We can find the key factors that influence the neuronal development, neuronal dynamics, if we are firstly able to observe and quantify the development. The reconstructed 3D line structure is one of the popular neuronal representations used in the study of neuronal development. Nevertheless, conventionally, every tiny detail of the alternation of the neuron at two different time points is labelled by human experts manually. This extremely postpone the progress of this study. In the second part of this dissertation, we present a supervised 4D neuronal Structural Plasticity Analysis (4D SPA) computer method that helps the researchers to explore theneuronal dynamics. The processing time for the user to complete the matching task was recorded; moreover, the time for two automatic methods to compute the matching results were recorded respectively, too. These records showed that with the help of 4D SPA, the consuming time for the exploration is greatly reduced, and furthermore the proposed method is more reliable.