C-arm影像輔助手術導航系統已廣泛的應用在骨科手術上，然而在C-arm前後向影像上很難正確規劃脊椎手術的路徑，因此系統使用上受到限制。本研究發展一套整合二維 C-arm影像及三維電腦斷層影像的方位校準方法，來獲得C-arm與CT影像座標系之間的轉換矩陣，使術前於CT影像上規畫好的路徑或是植入物模型經座標轉換能顯示在術中C-arm影像上，作為手術導引的依據。手術時，手術器械的方位也即時顯示在CT影像以及C-arm影像上，協助醫師更準確與可靠的定位手術器械。 首先將 C-arm影像扭正，並計算其X-ray發射源，然後選取 C-arm影像與 CT 影像上三個相同的特徵點，利用特徵點的座標建立C-arm與 CT影像座標系的初始方位校準，接著以光束投影法(Ray Casting)產生模擬X-ray成像原理的DRR影像。本研究採用Nvidia的CUDA開發環境來平行處理X-ray吸收性的線積分運算，以五種灰階為基礎的相似性量測(正規化交互關聯性、梯度關聯性、區域強度關聯性、梯度差異關聯性、共同資訊演算法)搭配三種最佳化方法(包威爾演算法、下坡單型法、基因演算法)進行比較，評估各種演算法在影像對位結果的準確性、收斂範圍與花費時間。結果顯示以正規化交互關聯性作為影像量測的目標函數，並使用下坡單型法搜尋最佳的方位校準矩陣，可使C-arm影像與DRR影像達到最大的相似度。 2D-3D方位校準誤差以脊椎假體模型進行實驗評估，初始收斂範圍限定在±10mm和±10度內，實驗共四十組，平均位移誤差及平均角度誤差分別為0.22mm與0.25 度，成功率為90%，平均對位所需時間為16秒。 C-arm image assisted surgical navigation system has been broadly applied to orthopedic surgery. For spinal surgery, accurate path planning on the C-arm AP image is difficult. Therefore, the applicability of the system is restricted. This research develops a 2D C-arm/3D CT image registration method to obtain the transformation matrix between C-arm and CT image coordinate frames. Through the transformation matrix, the preplanned surgical path or implant model on preoperative CT images can be transformed and displayed on the C-arm images for surgical guidance. During operation, the locations of surgical instruments will also be displayed on both CT and C-arm images to help the surgeon to precisely and safely position surgical instruments. First, the C-arm images are calibrated and the focus point of X-ray is determined. Then, select three identical characteristic points from C-arm images and CT images to obtain the initial registration between the C-arm and CT image frames. After that, the ray-casting algorithm is applied to generate digital reconstructed radiographs (DRR) from CT images. To speed up the generation of DRR, an Nvida’s CUDA graphics processing unit (GPU) is used for parallel computing of linear integration of X-ray absorptivity. Five similarity measures of 2D-3D registration including Normalized Cross-Correlation, Gradient Correlation, Pattern Intensity, Gradient Difference Correlation, and Mutual Information combined with three optimization methods including Powell, Downhill Simplex, and Genetic Algorithm are applied to evaluate the performance of converge range, efficiency and accuracy. The results show that the combination of Normalized Cross-Correlation measure with Downhill Simplex optimization algorithm has maximum correlation and similarity in C-arm and DRR images. Saw bone models are used in the experiment to evaluate registration accuracy. The initial convergence range is set within ±10mm and ±10 degree. The average errors in displacement and orientation of forty experiment sets are 0.22mm and 0.25° respectively. The success rate is approximately 90% and average registration time takes 16 seconds.
