Automatic Skull Reconstruction by Deep Learnable Symmetry Enforcement

TL;DR

An automatic skull reconstruction method based on the enforcement of skull symmetry using a learnable deep learning network that requires significantly fewer computational resources compared to other well-performing methods and is able to improve the reconstruction for the out-of-distribution cases.

Abstract

Every year, thousands of people suffer from skull damage and require personalized implants to fill the cranial cavity. Unfortunately, the waiting time for reconstruction surgery can extend to several weeks or even months, especially in less developed countries. One factor contributing to the extended waiting period is the intricate process of personalized implant modeling. Currently, the preparation of these implants by experienced biomechanical experts is both costly and time-consuming. Recent advances in artificial intelligence, especially in deep learning, offer promising potential for automating the process. However, deep learning-based cranial reconstruction faces several challenges: (i) the limited size of training datasets, (ii) the high resolution of the volumetric data, and (iii) significant data heterogeneity. In this work, we propose a novel approach to address these challenges by enhancing the reconstruction through learnable symmetry enforcement. We demonstrate that it is possible to train a neural network dedicated to calculating skull symmetry, which can be utilized either as an additional objective function during training or as a post-reconstruction objective during the refinement step. We quantitatively evaluate the proposed method using open SkullBreak and SkullFix datasets, and qualitatively using real clinical cases. The results indicate that the symmetry-preserving reconstruction network achieves considerably better outcomes compared to the baseline (0.94/0.94/1.31 vs 0.84/0.76/2.43 in terms of DSC, bDSC, and HD95). Moreover, the results are comparable to the best-performing methods while requiring significantly fewer computational resources (< 500 vs > 100,000 GPU hours). The proposed method is a considerable contribution to the field of applied artificial intelligence in medicine and is a step toward automatic cranial defect reconstruction in clinical practice.

Figures (4)

• Figure 1: Example of a damage in the frontal bone that cannot be reconstructed directly by reflecting about the symmetry axis because the defect breaks the skull symmetry resulting in missing data.
• Figure 2: An overview of the proposed method. The method starts with the segmentation-based volumetric reconstruction enhanced by the proposed symmetry network and loss. In the second step, the initial reconstruction is fine-tuned by iterative image registration using the symmetry loss as the objective function.
• Figure 3: Exemplary cases from the three datasets: (i) the SkullFix containing real skulls with synthetic defects in parietal bones, (ii) the SkullBreak consisting of real skulls with synthetic defects in all areas of skull, (iii) the real skulls with real, clinical defects.
• Figure 4: The influence of the symmetry enforcement on a real defective case. Note that the baseline is unable to generalize into a large real defect. Nevertheless, the symmetry enforcement considerably improves the reconstruction quality, leaving only a small hole between the frontal and parietal bones (the image registration-based refinement limited by the diffusive regularization could not deform the initial reconstruction further).