應用廣義霍夫的邊緣相似度於超音波影像與電腦斷層掃描影像對位研究

癌症病患於治療規劃到實際接受放射線治療需等候一段時間，首先專業的醫師先對病患做電腦斷層掃描(Computed Tomography, CT)，利用電腦斷層掃描影像找出腫瘤並進行治療，但實際治療時腫瘤位置可能會有所改變，或者是臨床上病患呼吸運動而影響到腫瘤治療規劃的位置，為了改善此點，透過超音波(Ultrasonic, US)獲得的即時影像資訊，以輔助醫療人員判斷治療與規劃的理想範圍。本文提出一種應用廣義霍夫於邊緣特徵的超音波與電腦斷層掃描影像對位技術，由於超音波影像具有雜訊多的特性，利用灰階形態學斷開和非等向性擴散做影像前處理，解決器官內部灰階值分佈不均勻與去除較大的非特徵之雜訊區域，再依照超音波探頭在三維空間中所對應角度與位置，提取出電腦斷層掃描影像二維切平面。然後擷取出超音波與電腦斷層掃描影像邊緣特徵，最後本文應用廣義霍夫的正規化相關係數法(Generalized Hough Normalized Cross Correlation, GHNCC)與正規化相關係數法(Normalized Cross Correlation, NCC)、邊緣正規化相關法(Edge Normalized Correlation ,ENC)、廣義霍夫轉換(Generalized Hough Transform, GHT)、矩不變量(Moment Invariant, MI)進行位置估測並將影像對位。本實驗使用CIRS公司出產的假體Model 071做為影像資料來進行五種不同影像對位演算法的測試，並且比較與分析演算法之間的精度與對位速度。

關鍵字

CT-US影像對位；邊緣相關；正規化相關；廣義霍夫；任意角度切面

並列摘要

In general, the delay from planning a course of treatment for cancer to the actual radiation therapy is over a month. In addition, the possibility of radiation therapy injuring healthy tissues will increase because the patient associates with breathing and movement in the treatment yielding the changed position of the tumor. The CT provided excellent information for spatial resolution, and doctors all use computer tomography to find tumors and to put their treatment plan into practice. In contrast, ultrasound can provide real-time information, and the capital and danger are also lower than other medical modalities. Combining spatial analysis and ultrasound to perform image registration will provide doctors with quicker and more accurate imagery and increase the effectiveness of radiation therapy. This thesis proposed a US and CT image registration technique based on Generalized Hough Transform applying the information of edge features and further compared it to different registration methods. Since the ultrasonic image contains complex noise, the gray-scale morphology opening and anisotropic diffusion filter were used to smooth the nonuniform intensity inside some specified organs and remove the noise. Then, the target slice of CT image was extracted according to the pose of US probe in three-dimensional space. Afterwards, the edge features were extracted from both US and CT images. Finally, Generalized Hough Normalized Cross Correlation (GHNCC), Normalized Cross Correlation (NCC), Edge Normalized Correlation (ENC), Generalized Hough Transform (GHT) and Moment Invariant (MI) were used to register the position. In the experiment, all the CT and US images were captured from the phantom Model 071 produced by CIRS company to test the performance of five image registration algorithms. The registration accuracy and efficiency of the above algorithms were compared and analyzed in detail.

並列關鍵字

CT-US Image registration ； Edge Normalized Correlation ； Normalized Cross Correlation ； Generalized Hough Transform ； any angle slice

參考文獻

[1] Brady Z., Taylor M.L., Haynes M., Whitaker M., Mullen A., Clews L., Partridge M., Hicks R.J., Trapp J.V., "The Clinical Application of PET/CT: A Contemporary Review, " Australas Phys Eng Sci Med ., 31(2): p.90-109, 2008.

[2] Yufei C., Zhicheng W., Weidong Z., Xiaochun Y., "Liver Segmentation from CT Images Based on Region Growing Method," ICBBE., art. no. 5163018, 2009.

[3] Zhang Z.,"Adaptive Region Intensity Based Rigid Ultrasound and CT Image Registration," CVPR., art. no.4587748, 2008.

[4] Karl K., Carl-Fredrik W., Ron K. and Kirby G.V., "Oriented Speckle Reducing Anisotropic Diffusion," IEEE TRANSACTIONS ON IMAGE PROCESSING , VOL. 16, NO. 5, MAY 2007.

[5] MacQueen J.B.,"Some Methods for Classification and Analysis of Multivariate Observations, " Proceedings of 5-th Berkeley Symposium on Mathematical Statistics and Probability, Berkeley, University of California Press, pp.281–297, 1967.

國際替代計量

應用廣義霍夫的邊緣相似度於超音波影像與電腦斷層掃描影像對位研究

未授權

主題瀏覽