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.

What problem does this paper attempt to address?

### Problems the paper attempts to solve This paper aims to solve several key challenges in the process of skull reconstruction: 1. **Limited size of the training data set**: Since it is difficult to obtain real - skull defect data, the existing training data sets are usually small, which limits the learning ability of the model. 2. **High - resolution volumetric data**: Skull reconstruction requires processing high - resolution three - dimensional image data, which leads to an increased demand for video random - access memory (VRAM). 3. **High heterogeneity of data**: Skull defects vary significantly in size, shape and thickness, which makes it difficult for the model to generalize to unseen data. ### Solutions To solve the above problems, the authors propose a skull reconstruction method based on learnable symmetry. Specifically, this method is achieved through the following steps: 1. **Symmetry Network (SN)**: Design a neural network to calculate the axis of symmetry of a healthy skull. This network can be used to enhance symmetry during the reconstruction process. 2. **Reconstruction Network (RN)**: Use the classic encoder - decoder architecture for defect reconstruction. 3. **Symmetry enhancement**: Introduce symmetry loss (SL) during the training process to ensure that the reconstructed skull is as symmetrical as possible. In addition, symmetry loss can also be used in the post - reconstruction refinement step to further optimize the reconstruction result through deformable image registration. ### Method overview - **Symmetry network**: Input a healthy skull and output the plane equation of the axis of symmetry. Perform unsupervised training through the symmetry loss function (SL) to calculate the axis of symmetry. - **Reconstruction network**: Input a skull with defects and output the initial reconstruction result. Combine the output of the symmetry network and further optimize the reconstruction result through reflection operations and the symmetry loss function. - **Refinement step**: After reconstruction, further optimize the reconstruction result through deformable image registration to ensure that the reconstructed skull is as symmetrical as possible. ### Experimental results - **Quantitative evaluation**: Quantitative evaluations were carried out on the SkullBreak and SkullFix data sets, and the results show that the symmetry - enhanced reconstruction method outperforms the baseline method in terms of DSC, bDSC and HD95 metrics. - **Qualitative evaluation**: Qualitative evaluations were carried out using real - life clinical cases, and the results indicate that symmetry enhancement significantly improves the reconstruction quality, especially when dealing with large defects. ### Conclusion This paper proposes an automatic skull reconstruction method based on symmetry enhancement. This method can significantly improve the reconstruction quality with relatively low computational resource requirements. This method represents an important progress in automatic skull defect reconstruction in clinical practice.

Automatic Skull Reconstruction by Deep Learnable Symmetry Enforcement

Skull shape reconstruction using cascaded convolutional networks

Automatic Cranial Defect Reconstruction with Self-Supervised Deep Deformable Masked Autoencoders

Improving Deep Learning-based Automatic Cranial Defect Reconstruction by Heavy Data Augmentation: From Image Registration to Latent Diffusion Models

Automatic Skull Defect Restoration and Cranial Implant Generation for Cranioplasty

Creating high-resolution 3D cranial implant geometry using deep learning techniques

Deep Learning-based Framework for Automatic Cranial Defect Reconstruction and Implant Modeling

CraNeXt: Automatic Reconstruction of Skull Implants With Skull Categorization Technique

Synthetic skull bone defects for automatic patient-specific craniofacial implant design

Three-dimensional deep learning to automatically generate cranial implant geometry

Cranial Implant Design Using a Deep Learning Method with Anatomical Regularization.

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

Cranial Implant Design via Virtual Craniectomy with Shape Priors

Automatic detection of symmetry plane for computer-aided surgical simulation in craniomaxillofacial surgery

Cranio-maxillofacial post-operative face prediction by deep spatial multiband VGG-NET CNN

Automated skull damage detection from assembled skull model using computer vision and machine learning

Cranial Implant Prediction using Low-Resolution 3D Shape Completion and High-Resolution 2D Refinement

Flip-avoiding interpolating surface registration for skull reconstruction

Skull-to-Face: Anatomy-Guided 3D Facial Reconstruction and Editing

A Baseline Approach for AutoImplant: the MICCAI 2020 Cranial Implant Design Challenge

Craniofacial Reconstruction Based on a Hierarchical Dense Deformable Model