三維醫學資料之光場顯示

Abstract

影像導引醫療處置具備侵入性低、準確度高、後遺症少等特性，近年來被廣泛應用於臨床；若將擴增實境技術與影像導引醫療處置結合，可以進一步提升手術表現。然而在觀看一般的擴增實境顯示器時，執刀醫師會受視覺輻輳調節衝突所苦而感受到暈眩與不適；所幸光場擴增實境顯示器不會造成視覺輻輳調節衝突，能夠提供最舒適的觀看體驗。在本論文中，我們將傳統的電腦斷層掃描資料轉換為光場，使得手術導引內容可以正確地呈現於光場擴增實境顯示器上。我們針對手術導引內容，搭配現有醫學影像顯示方式，設計一套轉移函數(transfer function)，使得由電腦斷層掃描影像生成之三維渲染內容，能順利呈現於光場擴增實境顯示器上，並允許醫生自由地調整導引內容的樣貌。此外，我們在考量到處理速度的重要性後，提出了一種適用於無結構體數據三維渲染的自適應性採樣方法，以減少採集的樣本數量；相較於均勻採樣方法，該方法能在更短的處理時間內，產生更高品質的渲染結果。最後，我們利用實驗室之前開發的近眼光場擴增實境眼鏡驗證所設計的光場生成方法的有效性。結果顯示，我們設計的方法可以有效率地將電腦斷層掃描資料轉換為光場，提供醫生高品質的手術導引內容。

Augmented reality (AR) technology has received considerable attention for image-guided medical procedures because it can be applied to make surgeries and therapies less invasive, more precise, and safer. However, due to the issue of vergence-accommodation conflict (VAC), conventional AR displays can easily cause visual discomfort or eyestrain to surgeons during medical procedures. Light field AR displays are free of the VAC and regarded as the ultimate display as they provide a natural and comfortable visual experience by reproducing light rays of the virtual object. In this thesis, we propose a systematic solution that enables traditional 3-D medical data to be converted to a format for a light field display. Our solution consists of a preprocessing module that converts tomographic data from the DICOM format to the 3-D volumetric data format and a volume rendering technique that efficiently generates high-quality light field content. To generate virtual object suitable for medical procedures, we propose a window transfer function that complies with the traditional display of medical data and allows users to freely adjust the contrast, brightness, and transparency to enhance the appearance of the organs under examination. To improve computational efficiency, we propose a novel adaptive sampling scheme to reduce the number of required samples for medical 3-D rendering. Our experiments quantitatively and qualitatively verify the effectiveness of the proposed adaptive sampling scheme, which performs favorably against the common linear sampling scheme. Finally, we verify the effectiveness of our proposed solution by displaying the resulting light fields on a near-eye light field display prototype. It shows that our solution to medical data conversion is suitable for real-world light field displays.