自動化粒線體之型態分析系統

An Automated Analysis System for Mitochondrial Morphological Variations

林俊志(Chun-Chih Lin)

指導教授 ： 蔡育秀

中原大學/工學院/生物醫學工程研究所/碩士(2010年)

https://doi.org/10.6840/cycu201001005

摘要

細胞生理學中，粒線體形態是呈現立體細管、小球狀與網狀的結構。而其經由持續不斷的融合及分裂來保持型態上的穩定並達到維持細胞能源供應的目的。當粒線體融合或分裂異常時，將導致粒線體形態及功能的變異，也會影響細胞本身多方面的功能障礙。所以，了解粒線體型態的變化將有助於新藥的開發與了解藥物反應的機制。
本研究目的是建立一套自動化型態的分析系統，以觀察粒線體在藥物反應後所引發的細胞凋亡現象。主要方法是將影像作前級處理後再合併分別取灰階值大於180的區塊影像與刪除灰階值低於10的兩種圖像，然後提取合成影像內具鑑別性的特徵參數，包含偏心度、圓度、長寬比等，利用K-Means Cluster建立系統分類法則。
為驗證系統成效，將4組不同基因制的粒線體影像(control, fis1, WT-mfn2, DN-mfn2)進行辨識。測試樣本共有包含169隻細胞83張影像。經分類後的型態共有15774個粒線體，系統型態辨識率平均82.25%。其結果顯示本系統可協助研究人員加快辨識粒線體細胞的型態變化，縮短新藥開發時程。

關鍵字

粒線體型態 ； K-Means Cluster

並列摘要

The cell physiology shows the mitochondria shapes are in 3-D network. tubules or small globules. Mitochondrial morphological stability is accomplished by continuous fusion and fission to maintain their energy supply function. Defects in above processes will lead to mitochondria variations in morphology and loss of function. Therefore, understanding mitochondrial morphological changes shall be important to new drugs screen and reaction mechanism.

The aim of this research is to establish an automated morphology analysis system for observing the mitochondria apoptosis after specific drug reaction. Firstly, an adaptive local threshold program, named Micro-P, is utilized to identify images whether contain fragmented globules or networks predominantly. These two groups of images were further processed by a dual-threshold morphological filter to segment the large and small objects in the cell images. The eccentricity, roundness, and aspect ratio of the objects were extracted as the classification features. At last, the K-Means cluster classification paradigm is used to distinguish the morphological subtypes.

Eighty three images that contain 169 cells are used to test the system performance. The cells include the control, WT-mfn2, DN-mfn2 and Fis-1, four groups. Totally, 15774 mitochondria in these cells are classified, and the contents of mitochondrial morphological subtypes in individual cells are measured. The statistical results shown the system yield average 82.25% correctness. This indicated that the system is capable to assist researcher in mitochondrial subtype classification and , hence, to shorten the new drug development timeline.

並列關鍵字

Mitochondria morphology ； K-Means Cluster

參考文獻

[11]古添全，“活體細胞中囊泡蛋白質動態特性之自動化定量與定性分析系統”，中原大學醫工系研究所碩士論文，2008。

[15] 李玫憶，“細胞次結構影像辨識系統”，中原大學醫工系研究所碩士論文，2005。

[4] John Lafferty, Andrew McCallum, Fernando Pereira, “Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data”, Proceedings of the Eighteenth International Conference on Machine Learning Pages : 282 - 289, 2001.

[5] Xiang Chen and Robert F. Murphy, “Objective Clustering of Proteins Based on Subcellular Location Patterns”, Journal of Biomedicine and Biotechnology ,

[6] S.V. N. Vishwanathan, Nicol N. Schraudolph, Mark W. Schmidt, Kevin P. Murphy, “Accelerated Training of Conditional Random Fields with Stochastic Gradient Methods”, Pittsburgh, Pennsylvania Pages: 969 - 976 ,2006.