Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Marek Wodzinski, Daria Hemmerling, Mateusz Daniol

TL;DR

The paper tackles automatic cranial defect reconstruction for personalized implants without requiring extensive ground-truth defect annotations. It proposes a self-supervised deformable masked autoencoder (MAE) that masks random patches of healthy skulls, applies deformable augmentation to create heterogeneous defects, and learns to reconstruct full skull shapes using a Residual 3-D UNet. The approach yields quantitative and qualitative improvements over multiple state-of-the-art segmentation models on SkullBreak and SkullFix, while enabling real-time inference. This reduces dependence on annotated defect data and enhances generalizability across datasets, potentially accelerating and democratizing the production of patient-specific cranial implants.

Abstract

Thousands of people suffer from cranial injuries every year. They require personalized implants that need to be designed and manufactured before the reconstruction surgery. The manual design is expensive and time-consuming leading to searching for algorithms whose goal is to automatize the process. The problem can be formulated as volumetric shape completion and solved by deep neural networks dedicated to supervised image segmentation. However, such an approach requires annotating the ground-truth defects which is costly and time-consuming. Usually, the process is replaced with synthetic defect generation. However, even the synthetic ground-truth generation is time-consuming and limits the data heterogeneity, thus the deep models' generalizability. In our work, we propose an alternative and simple approach to use a self-supervised masked autoencoder to solve the problem. This approach by design increases the heterogeneity of the training set and can be seen as a form of data augmentation. We compare the proposed method with several state-of-the-art deep neural networks and show both the quantitative and qualitative improvement on the SkullBreak and SkullFix datasets. The proposed method can be used to efficiently reconstruct the cranial defects in real time.

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

TL;DR

The paper tackles automatic cranial defect reconstruction for personalized implants without requiring extensive ground-truth defect annotations. It proposes a self-supervised deformable masked autoencoder (MAE) that masks random patches of healthy skulls, applies deformable augmentation to create heterogeneous defects, and learns to reconstruct full skull shapes using a Residual 3-D UNet. The approach yields quantitative and qualitative improvements over multiple state-of-the-art segmentation models on SkullBreak and SkullFix, while enabling real-time inference. This reduces dependence on annotated defect data and enhances generalizability across datasets, potentially accelerating and democratizing the production of patient-specific cranial implants.

Abstract

Thousands of people suffer from cranial injuries every year. They require personalized implants that need to be designed and manufactured before the reconstruction surgery. The manual design is expensive and time-consuming leading to searching for algorithms whose goal is to automatize the process. The problem can be formulated as volumetric shape completion and solved by deep neural networks dedicated to supervised image segmentation. However, such an approach requires annotating the ground-truth defects which is costly and time-consuming. Usually, the process is replaced with synthetic defect generation. However, even the synthetic ground-truth generation is time-consuming and limits the data heterogeneity, thus the deep models' generalizability. In our work, we propose an alternative and simple approach to use a self-supervised masked autoencoder to solve the problem. This approach by design increases the heterogeneity of the training set and can be seen as a form of data augmentation. We compare the proposed method with several state-of-the-art deep neural networks and show both the quantitative and qualitative improvement on the SkullBreak and SkullFix datasets. The proposed method can be used to efficiently reconstruct the cranial defects in real time.
Paper Structure (6 sections, 4 figures, 1 table)

This paper contains 6 sections, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Deformable Masked Autoencoder
  4. Datasets and Experimental Setup
  5. Results and Discussion
  6. Conclusions

Figures (4)

  • Figure 1: The pipeline of the proposed deep deformable masked autoencoder. Firstly, The healthy skull is loaded, then masking patches are randomly selected, deformably transformed, and used for masking to produce the defective skull. Then, a dedicated encoder-decoder network is used to perform the reconstruction.
  • Figure 2: Exemplary cases from the SkullFix and SkullBreak datasets. Note the sharp boundaries in the SkullFix dataset, and the smooth, irregular ones in the SkullBreak dataset.
  • Figure 3: Exemplary visual comparison of the proposed deformable MAE to the volumetric segmentation methods.
  • Figure 4: The qualitative results of the proposed approach compared to the state-of-the-art segmentation networks. The "D" denotes the proposed approach with the deformable elastic deformations and "ND" is the masking with sharp edges. Note the significant improvement of the proposed approach for the SkullBreak dataset.