特徵表面匹配法應用於可形變的影像校準

Abstract

本論文提出特徵表面匹配法，用以匹配兩個無標記器官表面的點對點關係，這個表面點對應關係可以形成表面點位移，來實現可變形的圖像校準。外科醫師通常在手術前利用斷層掃描、核磁共振或超音波來擷取器官影像，從中收集器官的腫瘤、血管位置，進而產生腫瘤切除計畫。在微創手術過程中，外科醫師使用腹腔鏡取得術中器官表面資訊，並根據手術計畫來判斷腫瘤、血管的位置，但是在手術過程中器官可以被抬起、移動、捏擠、翻轉或翻轉，這些手術上的必要操作可能導致器官嚴重的變形，使得手術醫師很難根據器官表面資訊來精準地推測出腫瘤、血管的位置，因此切除腫瘤的困難度會提升，甚至會有切到大血管的風險。本論文提出特徵表面匹配法搭配有限元素分析法來計算腫瘤、血管的位置，進而避免上述問題。此方法構建了生物力學體積模型，並使用一種新穎的表面匹配方法來判斷術前和術中器官表面點對點對應關係，此對應關係可形成表面點位移，再利用有限元素分析法將術前的生物力學體積模型根據表面點位移來進行形變計算，進而得到術中血管、腫瘤的位置。本論文的驗證方法是使用目標校準誤差來評估準確性，離體豬肝的驗證結果顯示，內部標記誤差（代表腫瘤、血管的位置）為4.54±3.55公厘，表面標記的誤差為2.98±1.09公厘。本論文亦利用公開的肺臟數據來驗證，使用DIR-LAB的兩組肺臟數據進行的驗證，初始誤差分別為3.91±2.82公厘和11.77±7.12公厘，經過特徵表面匹配法計算表面點位移與有限元素分析法計算形變，可使目標校準誤差分別降至1.88±1.16公厘和4.77±2.59公厘。本論文也使用POPI的其中一組肺臟數據進行驗證，初始誤差為11.66±6.23公厘，形變計算後目標校準誤差可降至4.12±2.22公厘。與文獻的比較，本論文所提的方法不論初始誤差大或小，目標校準誤差均優於文獻的平均之上，故本論文所提的特徵表面匹配法應用於可形變的影像校準經驗證後，具有可行性以及高度的準確性。

This dissertation devises a featured surface matching method to match the correspondence between two marker-less surface points for deformable image registration. Surgeons usually glean preoperative organ information, such as anatomy and the locations of tumors or large blood vessels, from the preoperative organ images obtained using computed tomography scans, magnetic resonance imaging, or ultrasound. This information forms an intervention plan before the organ resection surgery for removing the tumor. During minimally invasive surgery, the surgeon uses the laparoscope to obtain information about the intraoperative organ surface and identify the locations of tumors and vessels using the preoperative information. However, the organ can be lifted, shifted, flipped, squeezed, or turned over during surgery. These manual operations can lead to severe deformation, so it is challenging to identify intraoperative tumors or vessels’ location. It is also difficult to accurately remove a tumor while avoiding injury to large blood vessels. The removal of the tumors located in the posterior of an organ or close to large blood vessels may run the risk of injuring the tumor or the large blood vessels during resection. This dissertation proposes a featured surface matching method to identify intraoperative vessels or tumors’ locations to avoid the above-mentioned problem. The proposed method constructs the preoperative biomechanical volume model and uses a novel surface matching method to determine the displacement or correspondence between the preoperative and intraoperative surface points. The preoperative volume model is deformed by the finite element model in terms of the displacement so that it aligns with the intraoperative surface model and shows the location of intraoperative vessels and tumors. The experiments use the target registration error to assess the accuracy of the proposed method. The experiment results with an ex vivo porcine liver show that the error in the internal marker (which represents the location of the tumor and the vessel) is 4.54 ± 3.55 mm, and the error in the surface marker is 2.98 ± 1.09 mm. The validation also uses public available lung datasets to access the target registration error. The validation with two DIR-LAB lung datasets show the target registration errors of 1.88 ± 1.16 mm and 4.77 ± 2.59 mm, while initial errors are 3.91 ± 2.82 mm and 11.77 ± 7.12 mm, respectively; and the validation with POPI lung dataset which has the target registration error of 4.12 ± 2.22 mm, while the initial error is 11.66 ± 6.23 mm. The comparison among literature, the proposed method outperforms the accuracy regardless of small or large deformations. The validations demonstrate the feasibility and high degree of accuracy of the proposed method.