Automated Cranial Implant Design Through Volumetric Shape Learning: Benchmarking Deep Models on the SkullBreak and SkullFix Datasets

Abstract

The reconstruction of cranial defects plays a critical role in neurosurgery. Therefore, an accurate design of the cranial implant is critical for the restoration of skull functionality. However, the design process for the cranial implant involves a tedious process that relies on expert intervention. This paper develops an automated cranial implant design framework based on volumetric shape learning. The proposed method has been evaluated on two benchmark datasets, i.e., SkullBreak and SkullFix. These two datasets are considered standard for cranial reconstruction. The proposed method uses deep learningbased 3D volumetric shape learning for the reconstruction of cranial defects from incomplete CT scans. The proposed system simulates the defect, extracts volumetric features, and finally completes the shape to achieve accurate results. The performance evaluation shows the ability to achieve accurate results with a high Dice Similarity Coefficient (0.93), Intersection over Union (0.88), Mean Absolute Error (0.021), Root Mean Square Error (0.034), Hausdorff Distance (2.7mm), and Structural Similarity Index (0.95).