標題: 自動擷取並量化神經細胞影像的型態特徵
Automatic Extraction and Quantification of Morphological Features from Neuron Images
作者: 柯志錡
Ke, Chih-Chi
陳玲慧
何信瑩
Chen, Ling-Hwei
Ho, Shinn-Ying
多媒體工程研究所
關鍵字: 神經細胞影像分析;神經細胞之型態;神經軸追蹤;神經軸結構;神經軸分枝次序;neuron image;neuronal morphology;neurite tracing;neurite structure;neurite branch order
公開日期: 2013
摘要: 神經細胞型態的分析對於研究神經細胞的架構和功能之間的關係很重要。為了避免於大量神經細胞影像中費力且漫長的型態特徵分析,以電腦輔助的系統不可或缺。在神經細胞型態中,神經軸結構(如第一神經軸、第二神經軸)的辨認能表徵實驗調節後產生的差異。為了分析神經軸結構,基礎神經細胞型態偵測以及神經軸分枝次序決定是必需的。基礎神經細胞型態,如細胞本體、神經軸,的偵測已被廣泛研究,然而,神經軸分枝次序決定的研究僅有少數。商業軟體HCA-Vision是其中最傑出的研究之一。本研究發展出NeurphologyS,一個可從稀疏染色神經細胞之影像自動擷取並量化型態特徵的系統。提出之系統可自動化量化由五個子集組成的53個型態特徵,其中包含神經軸結構特徵。為了做到神經軸分枝次序決定,我們提出了一個基於規則的神經軸增長以及反向分枝優先順序決定的方法(RANBO)。RANBO包含兩個步驟:1) 基於區域神經軸角度規則的神經軸樹增長 2) 反向的神經軸分枝優先順序決定及分枝次序指派。兩個影像資料集被用以評估提出之系統,其一為經綠色螢光蛋白轉染之海馬迴神經細胞影像,另一為可在HCA-Vision網站上取得的Sez-6 knockout皮質神經細胞影像。由量化結果可知,提出之系統具備如同HCA-Vision的區分神經分枝微妙改變之能力;而提出之系統在神經軸結構的量化方面與HCA-Vision相較下表現良好。以MATLAB實作之NeurphologyS是開放源碼的且可無償取得。
Analysis of neuronal morphology is crucial for studying the relation between structure and function of neurons. For avoiding laborious and tedious manual analysis of morphological features of massive neuron images, a computer-aided system is indispensable. Out of neuronal morphologies, identification of neurite structure such as primary neurites or secondary neurites can characterize the differences after experimental modulation. For analysis of neurite structure, detection of basic neuronal morphologies and neurite branch order decision is necessary. Detection of basic neuronal morphology (such as soma or neurite) has been well studied, however, there are few studies that can achieve neurite branch order decision. The commercial software, HCA-Vision, is one of the most outstanding studies. This study develops NeurphologyS, a system to automatically extract and quantify morphological features from images of sparsely stained neurons. With the proposed system, 53 morphological features categorized into 5 subsets, can be quantified automatically. For neurite branch order decision, a rule-based neurite tree growing and backward branch priority decision (RANBO) is proposed. RANBO has two steps: 1) neurite tree growing based on local neurite angle rules 2) backward decision of neurite branch priority and branch order assignment. Two image datasets are utilized to evaluate the proposed system, one is image dataset of GFP transfected hippocampal neurons, and the other is the Sez-6 knockout cortical neuron image dataset available on HCA-Vision website. It can be known from our quantification results that the proposed system possesses the ability to differentiate subtle changes in neurite branching as HCA-Vision, and that the proposed system performs well compared to HCA-Vision in the quantification of neurite structure. The MATLAB-implemented NeurphologyS is open-source and freely available.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT070056638
http://hdl.handle.net/11536/72507
顯示於類別:畢業論文


文件中的檔案:

  1. 663801.pdf