FracFormer: Fracture Reduction Planning With Transformer-Based Shape Restoration and Fracture Data Simulation.

Area of Science:

• Orthopedic Surgery
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate orthopedic fracture reduction planning is critical for patient outcomes.
• Current automatic methods struggle with complex fracture geometries and limited annotated data.

Purpose of the Study:

• To develop a novel approach for automatic orthopedic fracture reduction planning.
• To overcome limitations of existing methods regarding geometric complexity and data scarcity.

Main Methods:

• Developed a transformer-based model for iterative fracture reduction pose refinement.
• Integrated learning-based shape restoration with a deformable fracture generation model (DFGM).
• DFGM utilizes statistical shape modeling and clinically representative patterns to generate synthetic data.

Main Results:

• Achieved mean errors of 1.85 mm and 3.40° on clinical data (hipbone, sacrum, femoral shaft fractures).
• Outperformed template-based and existing learning-based methods.
• Models trained on DFGM-synthesized data demonstrated strong generalizability to real clinical data.

Conclusions:

• The proposed method effectively addresses challenges in orthopedic fracture reduction planning.
• The integration of shape restoration, fracture simulation, and synthetic data generation shows significant promise.
• Validated by clinical and cadaver studies, the approach offers a robust solution for improved surgical planning.