Dual-plane X-ray–based reconstruction of bone geometry and ligament footprints using statistical shape modelling in the canine stifle

Abstract Accurate subject-specific bone geometry and ligament attachment sites are important for biomechanical analysis and surgical planning of the canine stifle joint. This study developed and validated a dual-plane X-ray–based 2D–3D reconstruction framework for simultaneous estimation of subject-specific bone geometry and ligament footprint locations. A footprint-embedded deformable shape template was constructed by integrating ligament footprint outlines into femoral and tibial statistical shape models derived from CT datasets. Bone reconstruction was performed using an alternating two-stage optimization combining point-to-plane correspondence–based pose estimation and silhouette-based shape refinement. The framework was validated using 12 canine hindlimb specimens with CT-derived models as ground truth. X-ray fluoroscopic images were acquired sequentially at different viewing angles and paired to form dual-plane X-ray configurations with angular separations of 90°, 60°, and 30° for evaluation. The root-mean-square errors for the reconstructed distal femur and proximal tibia ranged from 0.4 to 0.5 mm. The 90° and 60° configurations achieved significantly higher reconstruction accuracy than the 30° configuration. Mean ligament footprint centroid errors ranged from 1.6 to 3.1 mm across the ligaments and insertion sites. In conclusion, the proposed framework achieved submillimetre bone shape reconstruction errors with concurrent ligament footprint estimation under the evaluated ex vivo conditions when dual-plane X-rays were acquired with larger angular separations.