以X光顯微鏡進行三維斷層影像重建

Abstract

由於高解析度影像在取像上並不是那麼容易，所以從這些影像上取得的資訊是十分有價值的。X光顯微鏡利用了硬X光的短波長與高穿透力的特性來取得厚實標本的奈米解析度影像，這是其他取像技術無法辦到的。一般的X光取像設備是以X光吸收率來取像，但若以折射率來進行相位取像可有效地取得物體材質間的對比變化並加強之，且這種取像過程不會被某些生物標本的高X光吸收結構所影響。這種先進的技術在生醫研究上具有獨特的價值，但也浮現許多問題需要解決。本研究針對其中幾個問題並提出新的影像處理方法來解決。在斷層取像過程中，物體的旋轉會發生影像上的不對稱抖動現象，我們提出一個特徵點對齊法來對齊這些X光投影影像，使得斷層掃描影像可以成功的重建出來。為了解決高解析度攝影視野的限制，我們提出一個影像拼貼的技術將數張較小視野的影像合成一張可完整包含物體的大尺寸影像。此方法經由一組模擬資料來驗證其大型物體的三維斷層影像重建之可行性。最後，我們以傅立葉體積描繪法來快速地瀏覽重建完成的三維立體資料，也運用了具備美觀性與準確性的紋理體積描繪法來進行三維立體資料的視覺化。其中，我們為傅立葉體積描繪法提出一個快速的資料分類方法，可有效地改變轉換函數的權重並即時地反映在繪圖結果上。

High resolution images pushing the performance limit are always difficult to acquire and therefore extremely valuable to extract crucial information from them. X-ray microscopy takes advantage of the short wavelength and high penetration characteristic nature of hard-X-rays and demonstrated nanometer level resolution on thick specimens that is not available by other techniques. With devices to obtain contrast from differences of the refractive index, rather than the absorption of X-rays, the phase imaging can effectively enhance the contrast of materials which does not contain high X-ray absorbing structures such as biology specimens. The cutting edge level of performance found unique value in particularly biomedical research but also highlighted several problems which requires attention. This study addresses specifically several issues and tackles them by new image processing approaches. An image alignment method is implemented to eliminate the problem caused by the nonsystematic jitter of the sample rotation mechanism using a feature-recognition based alignment algorithm so a tomography reconstruction can be performed on these aligned projection images. To overcome the limitation of small field of view associated with the high resolution and limited number of imaging pixels, another image alignment method is developed to stitch these small field-of-view images into a larger one covering larger area of the specimen. This method is tested on a large number of simulated projection images of a phantom to demonstrate the possibility of tomography reconstruction in 3D of a large object. Finally, to visualize the tomography reconstructed images, a method based on the Fourier volume rendering (FVR) algorithm is designed to achieve better visualization and precision of the commonly used texture-based volume rendering (TVR) method. Improved processing efficiency by adjusting the weight of transfer function for FVR drastically reduces the processing time and computation resources required for voxel classification and make it possible for routine application.