- 學位論文
應用影像處理技術開發生物晶片影像分析系統
Application of Image Processing Technology to Develop Biochip Image Analysis System
摘要
在已開發出一套生物晶片影像分析系統，該系統使用96孔盤影像，影像本身會有扭曲問題，並且造成系統準確率不佳。故本研究開發一套生物晶片影像分析系統，可分析4×4及6×6之單孔盤微陣列晶片影像，除去影像扭曲之問題，並減少人工判讀影像時間，以提高晶片分析效率。 本系統把生物晶片影像輸入後將透過影像前處理、影像分割及特徵分類等步驟。首先，在影像前處理的步驟中，利用中值濾波去除雜訊並將影像平滑化，並對影像使用直方圖等化來增強影像對比度。接著，在影像分割的步驟中，使用二值化把反應點與背景分割出來，再利用圓的方程式切除孔盤外雜訊並分割出反應點。然後，在特徵分類的步驟裡把影像分為特徵點清晰與特徵點模糊兩類影像，特徵點清晰之二值化影像直接與特徵樣本比對並分析結果，特徵點模糊之影像則藉由原始影像之灰階值與特徵樣本比對並分析結果。最後分別使用假體影像來測試系統，並以191張Food晶片影像(4×4微陣列)、190張HPV晶片影像(6×6微陣列)及152張NTM晶片影像(6×6微陣列)進行系統之評估。 結果顯示，若光以反應點之觀測顯示，在Food菌體檢測晶片、HPV臨床檢體影像及NTM臨床檢體影像之準確率分別達到99.57%、99.50%及99.75%。若以單一孔盤之準確率統計，本系統之平均準確率為92.68%。在去除無效實驗結果後，三種晶片之陽性反應及陰性反應結果的準確率分別達到95.24%、96.81%及98.51%。 整體而言，本系統使用之單孔盤影像有效除去96孔盤影像的扭曲問題，並針對影像進行特徵分類，改善微陣列生物晶片影像分析系統準確率及影像判讀效率，對於生物晶片產業及使用者有正面的幫助。
並列摘要
In the previous study, we have developed a biochip image analysis system, which use the 96 plate hole image. There will be distortions in the image, and result in poor system accuracy. Therefore, this study developed a biochip image analysis system that analyzes 4×4 and 6×6 hole plate biochip images. It can be remove the problem of image distortion and reduces manual interpretation image time in order to improve the efficiency of the chip analysis. The biochip image was entered through this system and processed through the image pre-processing, image segmentation, and feature classification, three steps. First of all, in the image pre-processing step, median filter and histogram equalization are used to remove noise and smoothing the image and to enhance the image contrast. Secondly, the binarization method is used to separate the background and the reaction point for image segmentation. Then, using the circle equation to remove the noise outside of plate hole, and split the reaction point. Finally, the image is classifyed into clear feature point and blurs feature point. The bi-leveled image of clear feature point directly compared with the characteristics of the sample and analyzed the results. The image of blur feature point compared with the characteristics of the sample and analyzed the results by gray value of the original image. Designed pantom image is used to test and biochip images which including, 191 Food chip image (4×4 microarray), 190 HPV chip image (6×6 microarray) and 152 NTM chip image (6×6 microarray) are used to evaluate this system. The results shows that the order of the reaction point observations indicate the accuracy rate of food bacteria image, HPV clinical specimen image, and NTM clinical specimen image reached 99.57%, 99.50%, and 99.75%. In terms of the accuracy of a single hole plate, the average of the system's accuracy was 92.68%. After the removal of invalid responses, positive and negative results of the accuracy rates for three kinds of biochip image were 95.24%, 96.81%, and 98.51%, respectively. In conclusion, the system removed the distortion of 96 holes plate image by using single hole plate image, and carry on the feature classification for image. It can improve the accuracy of biochip image analysis system and the efficiency of the image interpretation which lead a positive assistance for biochip users and industry.
參考文獻
延伸閱讀
- 鄭伊斯（2012）。生物晶片影像分析系統之研發〔碩士論文，中原大學〕。華藝線上圖書館。https://doi.org/10.6840/cycu201200996
- 林汶正（2003）。以可程式微控制系統晶片為基礎之生理訊號監測系統〔碩士論文，中原大學〕。華藝線上圖書館。https://doi.org/10.6840/cycu200300610
- 丁鏞、張義鋒、陳一信（1998）。影像識別技術在晶粒取放機定位系統之應用。中原學報，26(1)，39-51。https://doi.org/10.6358/JCYU.199802.0039
- 楊仲準（2012）。應用於物理教學之影像數位分析技術。物理教育學刊，13(1)，41-49。https://doi.org/10.6212/CPE.2012.1301.05
- 蔡茵媜（2019）。Development of Simple and Rapid Enrichment Biochip〔碩士論文，中臺科技大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0099-0702202314391431