犬隻應用表面註冊於光學神經導航系統進行腦部採樣之可行性探討：大體實驗

Abstract

大腦結構精細且位於堅硬、不透明的顱骨內，使得臨床上對腦部進行活體組織之精準定位與採樣充滿挑戰。為提高手術的精確性與安全性，神經導航系統逐漸應用於臨床，其中光學式導航系統因操作靈活與應用範圍廣，日益成為主流。然而，目前獸醫領域多採用侵入性較高且操作複雜的標記點註冊法（fiducial-based registration），表面註冊法（surface-based registration）雖具有非侵入性及更簡化的操作流程，但在犬隻臨床應用之精準度尚未有系統性評估。因此，本研究旨在以犬隻大體模型，利用NaviVet®神經導航系統，評估光學神經導航系統應用表面註冊法進行腦部採樣的可行性，並與標記點註冊法進行比較，以提供獸醫臨床參考。 本研究使用8顆犬隻大體頭顱樣本。為進行術前規劃，首先於頭顱表面的左右顴骨弓、額竇中央偏側及枕骨隆突處植入四根骨釘後，進行電腦斷層掃描。於影像軟體中繪製導引目標後，匯入NaviVet®導航系統進行目標標記與路徑規劃。每顆頭顱左右大腦半球各規劃3–6對左右位置對稱的目標，共計72個目標（左右大腦半球各36個）。在註冊方法方面，右大腦半球採用表面註冊法，透過導航探針掃描頭顱表面擷取200個點，建立點雲模型完成註冊；隨後以NaviVet®系統導引脊髓針針芯至預先規劃的顱內目標，並以熱熔膠將其固定於顱骨表面。針芯固定後進行部分剪斷，末端保留外露。左大腦半球則採用標記點註冊法，利用先前植入的四根骨釘作為標記點完成註冊，並依相同流程導引針芯。所有針芯完成導引並經熱熔膠封固及部分剪斷後，進行術後電腦斷層掃描。在數據分析方面，透過齊疊合術前與術後影像，比對目標中心與針芯尖端座標，計算其距離（即目標註冊誤差，target registration error, TRE）作為導引精準度指標。此外，亦比較兩種註冊法之TRE、導引路徑長度、均方根誤差（root mean square error, RMSE）與操作時間，並分析可能影響導引精度之因素。 研究結果顯示，表面註冊之TRE（3.22±1.71 mm）與標記點註冊之TRE（3.74±1.75 mm）間無顯著差異（p = 0.158），兩種註冊法於導引精準度上表現相當。然而，表面註冊之操作時間（10.63±2.17分鐘）顯著短於標記點註冊（22.37±5.68分鐘；p < 0.05），展現其操作效率之優勢。此外，兩種註冊法之RMSE與TRE間僅呈低至中度正相關且未達顯著線性相關，暗示RMSE無法單獨作為導引精度預測指標。表面註冊的導引路徑長度與TRE呈顯著負相關（r = –0.358, p < 0.05），顯示路徑較短的目標反而可能伴隨較高的導引誤差。至於不同腦葉位置之TRE，無論何種註冊法，皆未呈現統計上顯著差異。 綜合上述結果，本研究證實犬隻使用光學神經導航系統於大體中進行腦部採樣時，應用表面註冊法之可行性；其精準度與傳統標記點註冊法相當，且展現操作簡便與低侵入性等優勢，具備良好之臨床推廣潛力。未來研究建議擴大樣本規模，增加病灶位置的多樣性，並評估更多元的表面掃描策略，以確認表面註冊技術在獸醫臨床實物中的可靠性與穩定性。

Due to the brain’s intricate structure within the rigid, opaque skill, accurate in vivo brain tissue localization and targeting remain challenging. Neuronavigation systems are increasingly being adopted in clinical practice to enhance surgical precision and safety, with optical tracking systems gaining popularity due to their procedural flexibility and board applicability. In veterinary neurosurgery, fiducial-based registration (FBR) is widely used. However, it is invasive and technically demanding. Surface-based registration (SBR) offers a non-invasive, simplified alternative, yet its accuracy in canine applications has not been evaluated. This study evaluated the feasibility and accuracy of SBR compared to FBR using the NaviVet® optical neuronavigation system in canine cadaver models. Eight canine cadaver heads were used. Four Steinmann pins were implanted at the bilateral zygomatic arches, the paramedian region of the frontal sinus, and the external occipital protuberance, followed by preoperative computed tomography (CT) scanning. Phantom targets were then created using imaging software and imported into the NaviVet® system for trajectory planning. For each head, 3–6 pairs of bilaterally symmetrical targets were created in the left and right cerebral hemispheres, resulting in a total of 72 targets in the study (36 pairs across hemispheres). SBR was performed for the targets in the right hemisphere by acquiring 200 surface points via a tracked probe for point cloud generation and registration. A spinal needle stylet was then navigated into each target under navigation. Once positioned, it was fixed to the skull surface with hot melt adhesive, then partially cut to leave the distal end exposed. In contrast, FBR was performed for the targets in the left hemisphere using the four implanted pins as fiducial markers, followed by navigating spinal needle stylet to corresponding targets using the same navigation protocol. After all the stylets had been navigated to their targets, fixed with hot melt adhesive and partially cut, postoperative CT scans were acquired and coregistered with preoperative images to calculate target registration error (TRE), defined as the Euclidean distance between planned target coordinates and actual stylet tip positions. RMSE (root mean square error), trajectory length, and registration time were also recorded and analyzed. The results revealed no significant difference in TRE between SBR (3.22 ± 1.71 mm) and FBR (3.74 ± 1.75 mm; p = 0.158), indicating comparable navigation accuracy. SBR required significantly less registration time than FBR (10.63 ± 2.17 vs. 22.37 ± 5.68 minutes; p < 0.05), highlighting its procedural efficiency. In both methods, RMSE and TRE were weakly to moderately correlated without statistical significance, suggesting that RMSE alone may not reliably predict navigation accuracy. For the SBR group, trajectory length was significantly and negatively correlated with TRE (r = –0.358, p < 0.05), indicating shorter trajectories may paradoxically lead to higher targeting errors. No significant differences in TRE were observed among different brain lobes in either method. In summary, this study is the first to validate SBR feasibility for intracranial targeting in canine cadaver heads using an optical neuronavigation system. SBR achieved accuracy comparable to FBR while offering advantages in simplicity and minimal invasiveness, supporting its potential for clinical application. Future studies with larger sample sizes, diverse lesion locations, and alternative scanning strategies are warranted to evaluate the clinical robustness of SBR in veterinary neurosurgery.