使用雙平面X光透視攝影圖像以可變形模板預測犬隻膝關節韌帶附著區域

Abstract

準確的骨骼外型與韌帶附著點資訊對於生物力學研究及韌帶疾病的個體化 解剖重建至關重要。雖然電腦斷層掃描（CT）為傳統上建立骨骼模型的標準方 法，但其應用受限於較高的成本、麻醉相關風險以及較高的輻射暴露量等因素。 此外，犬隻膝關節韌帶附著點相關資訊在獸醫文獻中極為稀少。為克服上述問題， 特別是在犬隻患者中，先前研究建立了一套基於統計形狀模型（SSM）之可變形 形狀模板（DST），能夠利用 CT 建構之骨骼模型精確預測犬隻膝關節韌帶的附著點。 本研究旨在開發一種新演算法，利用雙平面 X 光影像結合基於輪廓配對之 SSM 二維至三維（2D–3D）註冊方法，以重建個體化骨骼外型，並套用先前建立 之 DST 於註冊後的骨骼模型上進行韌帶附著點的預測。為驗證本研究提出的方 法，將 2D–3D 註冊所得之骨骼模型與 CT 所得之骨骼模型進行重建誤差比較， 並評估分別使用不同來源骨骼模型的 DST 之韌帶附著點預測準確性。 本研究使用手術用 C-arm 透視系統，對納入於先前 DST 建構研究之 12 具犬 隻後肢標本進行與頭尾方向（cranial–caudal）視角相夾 0°、30°、60°與 90°等角 度之 X 光影像拍攝。2D–3D 註冊以 90°–0°、90°–30°與 90°–60°之三組雙平面影 像組合進行。當 SSM 依據 X 光影像進行 2D–3D 註冊時，DST 中所嵌入之韌帶 附著點亦會相應變形（DSTreg ），以進行個體化韌帶附著點預測。預測之韌帶附 著點將進一步與實際量測資料及以 DST 與 CT 骨骼模型進行最佳化擬合後所得之預測結果（DSTopt ）進行比較，並分析影像組合角度變化對骨骼重建與韌帶附著點預測準確度之影響。 股骨與脛骨骨骼模型之平均均方根誤差分別為 0.7–0.9 毫米與 0.6–0.7 毫米， 而韌帶附著點之歐幾里得距離誤差範圍則為 0.9–2.7 毫米。整體而言，DST reg 所得之預測結果與 DST opt 相當，顯示本研究提出之方法為一種準確且更安全的替代方案，無須依賴 CT 即可取得個體化的犬隻膝關節韌帶附著位置。

Accurate bone shape and ligament footprint data are critical for biomechanical research and patient-specific anatomical reconstruction in cases of ligament disorders. While computed tomography (CT) has traditionally been the standard for acquiring bone shape models, its application is limited by high cost, anesthesia-related risks, and increased radiation exposure. On the other hand, ligament footprint data was scarcely reported in veterinary medicine. To address these limitations, especially in canine patients, a previous study introduced a deformable shape template (DST) based on a statistical shape model (SSM), which enabled precise prediction of canine stifle ligament footprints using CT-derived bone models. The present study aimed to develop an SSM-constrained silhouette-based two-dimensional to three-dimensional (2D–3D) registration method using dual-plane X-ray images to reproduce patient-specific bone shape models. The DST from the previous study was applied to the registered bone shapes in an attempt to estimate ligament footprints. To validate the proposed method, reconstruction errors of the 2D–3D registered bone shapes were compared to those of CT-derived bone models, and the accuracy of ligament footprint estimations using the DST-based bone shapes obtained from different sources was evaluated. Dual-plane X-rays were acquired from 12 hindlimb specimens at 90°, 60°, 30°, and 0° to the cranial-caudal perspective using a surgical C-arm fluoroscopy system. For 2D–3D registration, image pairs at 90°–0°, 90°–30°, and 90°–60° were utilized. As the SSM was registered to the paired X-ray images, the embedded ligament footprints within the DST were deformed accordingly (DSTreg). Predicted ligament footprint locations were compared to the ground truth data and to those estimated by optimally fitting the DST to CT-derived bone models (DSTopt). The influence of image pair angles on the accuracy of bone shape reconstruction and ligament footprint estimation was evaluated. The average root-mean-squared error for femoral and tibial bone shape reconstructions ranged from 0.7–0.9 mm and 0.6–0.7 mm, respectively. The Euclidean distance error for ligament footprint predictions ranged from 0.9–2.7 mm. Overall, the accuracy of DSTreg was comparable to that of DSTopt, suggesting that the proposed method offers a viable and safer alternative for acquiring accurate, patient-specific ligament footprint positions without the need for CT imaging.