Back to all papers

Reverse Reconstruction of the Patient-Specific Anatomy of the Knee: Deep Learning-Based Point Cloud Completion Model.

November 13, 2026pubmed logopapers

Authors

Kong L,Liu J,Kang H,Xu C,Song H,Wang F

Affiliations (6)

  • Department of Orthopaedic Surgery, Third Hospital of Hebei Medical University, Shijiazhuang, Hebei, China 050051. Electronic address: [email protected].
  • School of Computer Science and Technology, Beijing Institute of Technology, Beijing, China 100081. Electronic address: [email protected].
  • Department of Orthopaedic Surgery, Third Hospital of Hebei Medical University, Shijiazhuang, Hebei, China 050051. Electronic address: [email protected].
  • Department of Orthopaedic Surgery, Third Hospital of Hebei Medical University, Shijiazhuang, Hebei, China 050051. Electronic address: [email protected].
  • School of Computer Science and Technology, Beijing Institute of Technology, Beijing, China 100081. Electronic address: [email protected].
  • Department of Orthopaedic Surgery, Third Hospital of Hebei Medical University, Shijiazhuang, Hebei, China 050051. Electronic address: [email protected].

Abstract

This study aimed to validate the effectiveness of a deep learning-based point cloud completion method for the reverse reconstruction of the patient-specific anatomy of the knee bone morphology, thereby providing technical support for personalized diagnosis and treatment. A computed tomography (CT) database containing 140 normal knees was constructed. There were 10 clinically common defect patterns (7 femoral, 3 tibial) generated. Multiple deep learning methods were employed to achieve defect point cloud completion. Virtual testing was conducted to compare the morphological consistency (assessed by Chamfer Distance, Density-Aware Chamfer Distance, and F-score) and anatomical parameter consistency between the resulting and original intact models. The optimal point cloud completion method, based on the evaluation results, was selected for preliminary clinical feasibility testing. Consistency between reconstructed affected-side models and the intact contralateral knee models was assessed in patients who have unilateral post-traumatic osteoarthritis. In virtual testing, the CRA-PCN deep learning model yielded the best reconstruction performance. For all defect patterns, the reconstructed geometries attained a chamfer distance below 0.45 mm, while deviations of anatomical parameters were limited to within 0.5 mm for linear measurements and 0.2° for angular measurements. In the preliminary clinical feasibility testing, the reconstructed morphology exhibited a chamfer distance of approximately 1.2 mm compared to the contralateral side. The restoration accuracy of the joint line angle was superior in extensive defect patterns, with the optimal deviation being within 0.2° on the femoral side and approximately 0.7° on the tibial side. The CRA-PCN model enables reverse reconstruction of defective knee structures with satisfactory anatomical restoration. It provides quantitative anatomical references for personalized arthroplasty and aids identification of deformity origins in osteoarthritis (OA) patients, though further large-scale prospective research is required to verify its clinical value.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.