Back to the Roots: Reconstructing Large and Complex Cranial Defects using an Image-based Statistical Shape Model.

Area of Science:

• Medical Engineering
• Computational Anatomy
• Biomedical Imaging

Background:

• Designing patient-specific cranial implants for large defects is complex.
• Current automated methods using convolutional neural networks (CNNs) struggle with real-world, complex cases.

Purpose of the Study:

• To develop and evaluate a statistical shape model (SSM) for automated cranial implant design.
• To compare the performance of SSM against CNN-based methods for large and complex cranial defects.

Main Methods:

• A statistical shape model (SSM) was constructed using skull segmentation masks represented as binary voxel occupancy grids.
• The SSM was evaluated on cranial implant design datasets, including large and complex real-world defects.
• Performance was compared against existing CNN-based approaches.

Main Results:

• CNNs performed better on synthetic defects.
• The SSM demonstrated superior performance for large, complex, and real-world cranial defects.
• Implants designed by SSM were deemed clinically feasible by neurosurgeons after minor adjustments.

Conclusions:

• Statistical shape models offer a viable alternative to CNNs for complex cranial implant design.
• The developed SSM shows potential for clinical application in cranioplasty.
• Publicly available datasets and model encourage further research and development.