視網膜影像視神經盤定位與血管架構擷取系統

A Novel Tracking System for Locating Optic Disk and Extracting Vessel Structure in Retinal Images

吳政聰(Cheng-Tsung Wu)

指導教授 ： 陳文雄

共同指導教授 ： 姜文忠(Wen-Chung Chiang)

國立暨南國際大學/科技學院/通訊工程研究所/碩士(2008年)

https://doi.org/10.6837/NCNU.2008.00055

摘要

視網膜影像廣泛地被使用在各種疾病的臨床診斷和治療上，透過對糖尿病患視網膜影像中血管圖像的觀察及偵測出可能的微血管變化，是一個重要的病變研判方式。在視網膜影像內血管進行研判和量度能有效完成對臨床病兆特徵進行定量評估，並可針對糖尿病患在視網膜病變方面的治療進行早期診斷和有效監控。本論文提出一套視神經盤與視網膜血管定位系統，利用簡單的最大投影量及數位影像處理的方法來完成定位任務。這套系統分為兩部份，分別為視神經盤定位與視網膜血管萃取模組。視神經盤定位模組先利用最大投影量的方法切割出一個包含完整視神經盤影像，減少背景雜訊影響定位誤差，再利用最小H轉換（H-minima transform）提高視神經盤與背景的對比，最後再利用最大投影量計算出視神經盤的圓心與半徑。在視網膜血管萃取模組中，先用多尺度技術，萃取出血管的梯度變化量與主要曲率這兩個血管特徵。利用這兩者特徵對原始影像血管部位做強化，提高血管與眼球組織之間的對比，再利用二值化方法定位出視網膜血管位置。在血管定位部份，本論文由台中榮總醫院資料庫（TCVGH）與公開資料庫DRIVE中收集了67張影像來測試所提出的方法。實驗結果顯示，在DRIVE資料庫中，定位結果達到84.12%正確接受率（true positive rate, TPR），3.82%錯誤接受率（false positive rate, FPR）,與81.9%準確性（accuracy, AC）。在榮總資料庫中，定位結果達到85.61%正確接受率，3.17%錯誤接受率,與82.8%準確性。最後本論文針對實驗的數據結果進行分析比較，來驗證建構本系統時所提的相關推論，以供後續研究做為參考。

關鍵字

視網膜影像 ； 視神經盤定位 ； 血管萃取

並列摘要

The retinal images are widely used in diagnosis and treatment of various diseases in clinics. An important aspect of diabetic retinopathy (DR) is the micro-vascular changes that cause detectable changes in the appearance of retinal blood vessels. Identification and measurement of blood vessels in retinal images could allow quantitative evaluation of clinical features, which may allow early diagnosis and effective monitoring of therapies in retinopathy. This thesis presents a retina localization system of optic disk (OD) and vessels. We will use simple techniques which are maximum project and digital image processing to complete this task. The propose system consists of two modules: OD location module and vessel extraction module. In OD location module, we segment complete OD image by using maximum projection in order to reduce locating error suffering from background noise. Then, we use H-minima transform to enhance the contrast between OD and background. Finally, we find the center and radius of OD by using maximum projection again. In vessel extraction module, we use multi-scale technique to extract two vessel features, which are the gradient magnitude and principal curvature of retinal vessel. Using these two vessel features can enhance the contrast between retinal vessel and background. Finally, we locate retina by binarization. In this thesis, 67 images are collected from Taichung Veterans General Hospital (TCVGH) database and a public database, named DRIVE. In DRIVE database, the experimental results show true positive rate (TPR) of 84.12%, false positive rate (FPR) of 3.82%, and accuracy (AC) of 81.9%. In TCVGH database, the experimental results show TPR of 85.61%, FPR of 3.17%, and AC of 82.8%. This thesis also analyzes the experimental results and provides useful information for future research.

並列關鍵字

Retinal Image ； Optic Disk Localization ； Blood Vessel Extraction.

參考文獻

[1] “http://www.tzuchi.com.tw/file/divintro/eye/instr/allegretto/HRA2.htm,” 慈濟眼科.

[2] “http://www.biometricgroup.com/,” Homepage of International Biometric Group.

[3] H. Li and O. Chutatape, “Automatic location of optic disk in retinal images,” in Proc. IEEE-ICIP, vol. 2, pp. 837-840, Oct. 2001.

[4] H. Li and O. Chutatape, “Automated feature extraction in color retinal images by a model based approach,” IEEE Trans. on Biomedical Engineering, vol. 51, No. 2, pp. 246-254, Feb. 2004.

[5] J. Lowell, A. Hunter, D. Steel, A. Basu, R. Ryder and E. Fletcher, “Optic nerve head segmentation,” IEEE Trans. on Medical Imaging, vol. 23, No. 2, pp. 256-264, Feb. 2004.