微晶片能夠在單張微晶片影像同時檢驗數千至數萬顆基因，這項技術被廣泛的應用於分子生物實驗上。然而，目前計算微晶片影像上的每一點的特徵值仍仰賴人工操作與手動調整，這不但耗時而且不具重現性。並且，一般商用微晶片分析軟體與生物公司都不透漏分析方法的細節，分析方法的不明確影響實驗結果的可靠度。在這篇論文我們提出一個方法以接近全自動的方法，計算基因的表現量，並由其中篩選出，在正常與腫瘤有不同表現量的基因，這些基因值得作進一步臨床檢驗。我們提出的方法首先將微晶片影像作灰階化，使每一點的量值代表基因的表現量，接著濾除實驗上殘留的物質所造成的雜訊。因為晶片掃描時會有角度誤差，造成定位圓點的錯誤，所以我們提出方法將影像旋轉至正確角度。生物晶片上的每一圓點，代表實驗所得的基因表現量，利用影像辨識的方法，定位每一點在晶片上的圓點，並將之從背景上分離出。最後計算每一圓點的統計數據作為特徵值。根據這些特徵值再作進一步的基因篩選。

我們由台大血管新生研究中心提供的26對正常與胃癌腫瘤組織製成的52片微晶片影像，利用我們提出的方法計算出每一片微晶片的特徵值。由這些特徵值，我們使用三種評估函數，篩選其中15顆共同具由差異表現量的基因，提供醫療人員作進一步研究。

DNA microarray technology is extensively adopted in molecular biology experiments owing to its ability to monitor thousands of genes simultaneously on a single microarray image. However, computing spot features from a microarray image still relies on manual operations and adjustments, which is occasionally inefficient and not repeatable. Moreover, the details of microarray analysis by commercial software and bio-company are usually not revealed. The unclarity of analysis methods incurs the discredit of the experiment correctness. This thesis proposes an approach for measuring gene expression level nearly automatically and identifying potential genes for further clinical investigation. First, the cDNA microarray image is processed by graying, smoothing and rotating. Next, each spot of the image is located and split from background. The spot statistics are finally computed as features. Based on these features, the potential genes are identified according to their separation.

Our approach is tested for computing spot features on 52 cDNA microarray images made from 26 pairs of normal and tumor tissues of patients of gastric cancer provided by the angiogenesis research center at National Taiwan University (ARCNTU) [13] and selects 15 genes for further clinical investigation.

[1] N. Brändle, H. Lapp, and H. Bischof. Fully Automatic Grid Fitting for Genetic Spot Array Images Containing Reference Spots. Technical Report PRIP-TR-058, PRIP, TU Wien, 1999.
[2] A. Brazma and etc., “Minimum Information about Microarray Experiment (MIAME) - toward Standards for Micorarray Data”, Nature Genetics, vol. 29, pp. 365-371, Dec. 2001.
[3] H.C. Causton, J. Quackenbush, and A. Brazma, Microarray Gene Expression Analysis, Blackwell Publishing, 2003.
[4] C.C. Chen and R.C. Dubes, Environmental Studies of ICM Segmentation Algorithm, Journal of Information Science and Engineering, vol. 6, pp. 325-337, Nov. 1990.
[5] Yidong Chen, Vishnu Kamat, Edward R. Dougherty, Michael L. Bittner, Paul S. Meltzer, Jeffery M. Trent: Ratio statistics of gene expression levels and applications to microarray data analysis. Bioinformatics 18(9): 1207-1215 (2002)
[6] Han-Yu Chuang, Hongfang Liu, Stuart Brown, Cameron McMunn-Coffran, Cheng-Yan Kao, and D. Frank Hsu, “Identifying Significant Genes from Microarray Data”, BIBE 2004
[7] R.A. Fisher, “The Use of Multiple Measurements in Taxonomic Problems,” Annals of Eugenics, vol. 7, part II, pp. 179-188, 1936.
[8] R.C. Gonzalez and R.E. Woods, Digital Image Processing, Prentice-Hall, Inc., 2002.
[9] Jay L. Devore. “Probability and statistics for Engineering and the Sciences”. Duxbury 1995
[10] Anil K. Jain. Fundamentals of Digital Image Processing. Prentice-Hall, 1986
[11] Khan et al., “Classification and Diagnostic Prediction of Cancers Using Gene Expression Profiling and Artificial Neural Networks,” Nature Medicine 7, no. 6 (2001) 673–679.
[12] N. Otsu. "A Threshold Selection Method from Gray-Level Histograms," IEEE Trans. on Systems, Man, and Cybernetics, vol. SMC-9, pp. 62-66, Sept. 1979.
[13] http://www.angio.bioinfo.ntu.edu.tw
[14] http://www.mediacy.com/arraypro.htm
[15] http://www.axon.com/GenePixSoftware.htm