摘要
本研究發展生物晶片影像分析系統，可分析5×5和6×6之微陣列晶片影像，其目的在於減少人工判讀影像所需時間，提高生物晶片分析效率。 本系統首先把生物晶片影像由彩色影像轉為灰階影像，利用Radon轉換將多孔盤晶片切割為單一孔盤，接著使用Sobel把影像邊緣強化並將影像二值化，並利用圓方程式擷取二值化影像中反應點，最後系統所擷取之反應點與特徵樣本比對並分析其結果。此外晶片的特徵種類可依照各實驗設定不同的特徵點，並建立樣本資料庫與擷取到的晶片影像比對分析結果。建構假體影像進行效能測試，並使用50張5×5和30張6×6微陣列影像作為系統驗證。 結果顯示：系統偵測範圍為反應點與背景的灰階值差大於25以上，且直徑像素大於9以上。於實際影像測試中，系統準確度為90%，其中5×5微陣列影像準確度92%；6×6微陣列影像準確度為87%。另外對於微陣列影像中的每個反應點之靈敏度為98.36%、專一性100%、準確度99.61%以及Kappa值98.92。 為增進系統之效能，未來系統將分為兩階段，第一階段先將影像分為清楚與模糊影像兩種，清楚影像透過系統自動判讀之準確度可達100%，第二階段則是針對模糊影像再進行其他影像處理之方法，或是提供影像資訊給使用者作為人工判讀之依據。
並列摘要
The purpose of this study is to develop an image analysis system for 5×5 and/or 6×6 microarray biochip which is can reduce time cost of manual interpretation image and improve the efficiency of analysis. Several image processing methods were used to develop system. First, this system changed biochip image from color image into grayscale image. The multi well biochip image was segmented to a single well by Radon transform. Then, the Sobel masker was used to enhance the image edge and then the image was bi-leveled. In addition, using circle equation, reaction point was gotten from bi-leveled image. Final, reaction point which captured from system was compared with pre-set characteristic pattern and analyzed its results. Moreover, in this system, feature type of chip could change by each experimental setting feature spot, and establish the sample database to compare with captured biochip image. The phantom images were constructed for testing the limitation of this system. And 50 pieces of 5×5 and 30 pieces of 6×6 microarray image were used for our system verification. Results show that system minimum detection range for the gray value difference between the reaction point and background is 25, and the size of pixels is diameter greater than 9. In real data, the system accuracy was 90%, in which 5×5 microarray images were 92% and 6×6 microarray images were 87%. Furthermore, the sensitivity of reaction point in microarray images was 98.36%, specificity was 100%, accuracy was 99.61% and Kappa value was 98.92. In order to improve this system, two analysis stages is proposed in the future. First stage, images will divide into clear and unclear type. Through the automatic interpretation, the clear images will improve accuracy up to 100%. The second stage, the other image processing methods will be used for unclear images or image information will be provided for user as a base of manual interpretation.