應用病患專屬數位分身模擬髕骨運動軌跡： 外側支持帶鬆解術之影響

Abstract

本研究建立一套病患專屬的膝關節數位分身系統，用以預測髕股關節在韌帶鬆解術後的運動反應，並著重於外側支持帶鬆解。此數位分身系統整合了新穎的二維至三維重建方法、統計形狀建模（SSM）、有限元素（FE）分析以及機器學習，旨在提升針對髕股關節不穩定與運動異常（maltracking）的手術技術。子集式配準（SBR）為一種新穎方法，可將三維模型與已標註之二維X光影像對齊。本研究結合SBR與SSM形變技術，以重建病患專屬的三維解剖模型。並以屍體實驗作為驗證依據。研究結果顯示，針對性鬆解比起擴大鬆解範圍更能有效改善運動異常。在膝關節彎曲小於60°時，對上外側支持帶（LR）進行約25%的部分鬆解，所達成的髕骨偏移修正效果幾乎與正常狀況一致；反之，進行較大範圍或完全（75–100%）的LR鬆解則導致過度修正。若同時搭配施行25–50%的內側支持帶（MR）鬆解，則可進一步改善髕骨的偏移與傾斜校正。屍體實驗亦支持有限元素模擬的預測結果，顯示若進行雙側完全鬆解，會導致髕骨傾斜活動度增加，指出關節不穩定的風險。然而，髕骨偏移的量測結果與正常運動軌跡吻合度不佳，主因可能為骨標記的遺失所致。在未進行鬆解的情況下，髕骨屈曲延遲達30–50%；在LR鬆解情境下則達60–75%，皆顯著低於常見的正常值，顯示實驗中存在誤差。在有限元素分析中，LEPL束，尤其是上外側髁髕韌帶（sLEPL），對於恢復正常髕骨傾斜具有良好效果。機器學習分析亦指出，sLEPL是影響髕骨傾斜與偏移最具影響力的參數，具有應用於針對性鬆解決策輔助的潛力。本研究成功建立並驗證了病患專屬數位分身系統中各個獨立模組於髕骨運動模擬中的可行性，提供術前外側支持帶鬆解決策之輔助依據。未來若能納入更多術前與術後的臨床資料，將有助於提升此數位分身系統之普遍適用性。

This study created a patient-specific digital twin system of the knee joint to predict the kinematic response of the patellofemoral joint following ligament release, focusing on lateral retinacular release. The digital twin aimed to improve the surgical technique for patellofemoral instability and maltracking by integrating a novel 2D–3D reconstruction method, statistical shape modeling (SSM), finite element (FE) analysis, and machine learning. Subset-based registration (SBR) is a novel method to align 3D models to annotated 2D X-rays, used here with SSM deformation to reconstruct a patient-specific 3D model. A cadaver study served as a validation benchmark. Results showed that targeted release corrected maltracking better than increasing release extent. Partial (~25%) release of the superior lateral retinaculum (LR) achieved shift correction nearly identical to normal below 60° flexion. Larger or full LR release (75–100%) caused over-correction. Combining 25–50% medial retinacular (MR) release with LR release improved shift and tilt correction. Cadaver study results supported FE predictions in that full bilateral release increased patellar tilt mobility, indicating instability. Shift measurements poorly matched normal tracking due to bone marker loss. Patellar flexion lagged 30–50% in no-release and 60–75% in LR release—lower than commonly observed normal values, indicating experimental inaccuracies. The LEPL bundle, particularly the superior lateral epicondylopatellar ligament (sLEPL), showed good restoration of normal patellar tilt in the FE analysis. Machine learning analysis also identified sLEPL as the most influential parameter affecting patellar tilt and shift—useful for targeted release decision support. This study successfully developed and validated the individual systems of the patient-specific digital twin system for patellar tracking for preoperative lateral retinacular release decision support. Large clinical datasets of preoperative and postoperative data could improve the generalizability of the digital twin system.