A shape-partitioned statistical shape model for highly deformed femurs using X-ray images

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computer-Aided Surgery

Background:

• Patient-specific 3D models are crucial for surgical planning and medical device design.
• Conventional statistical shape models (SSMs) have limitations in reconstructing complex anatomical variations.
• Existing SSMs may fail when target shapes fall outside the training data's variation range.

Purpose of the Study:

• To introduce a shape-partitioned statistical shape model (SPSSM) for enhanced patient-specific 3D femur reconstruction.
• To improve the accuracy of 3D femur models derived from X-ray images, especially for shapes with significant variations.
• To overcome the limitations of conventional SSMs in representing complex anatomical deformations.

Main Methods:

• Developed an SPSSM by segmenting femur shapes into anatomical regions.
• Created independent SSMs for each segment, preserving relevant variations from the original eigenvector matrix.
• Validated the SPSSM using femur deformation models (anteversion angle, femoral head scaling) and a leave-one-out cross-validation approach with twelve femurs.

Main Results:

• The SPSSM demonstrated reduced average reconstruction errors compared to the conventional SSM across all tested deformation models.
• For a 30° femoral head rotation, SPSSM error decreased from 5.34 mm to 3.82 mm.
• For femoral head size changes (±20%), SPSSM errors were reduced from 4.70 mm to 3.56 mm and 4.28 mm to 3.10 mm, respectively.

Conclusions:

• The proposed SPSSM significantly outperforms conventional SSMs in reconstructing patient-specific 3D femur models with complex variations.
• SPSSM offers a more robust approach for creating accurate 3D anatomical models from medical imaging data.
• This method has potential applications in improving orthopedic surgery planning and implant design.