Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

High-Resolution Cranial Defect Reconstruction by Iterative, Low-Resolution, Point Cloud Completion Transformers

Marek Wodzinski, Mateusz Daniol, Daria Hemmerling, Miroslaw Socha

TL;DR

This work reframes cranial defect reconstruction as a large-scale point-cloud completion problem and introduces an iterative, geometry-aware transformer to complete missing skull regions directly in 3-D. By replacing volumetric approaches with a sparse point-cloud representation and a DACD-based loss, the method achieves competitive reconstruction quality while dramatically reducing GPU memory usage, enabling high-resolution completion at practical speeds. The approach is validated on SkullBreak and SkullFix datasets, showing strong generalization and robustness to unseen cases, with ablations illustrating the benefits of iterative refinement and density-aware training. The findings suggest substantial practical impact for automating personalized cranial implants, reducing computational burden, and facilitating downstream 3-D printing workflows.

Abstract

Each year thousands of people suffer from various types of cranial injuries and require personalized implants whose manual design is expensive and time-consuming. Therefore, an automatic, dedicated system to increase the availability of personalized cranial reconstruction is highly desirable. The problem of the automatic cranial defect reconstruction can be formulated as the shape completion task and solved using dedicated deep networks. Currently, the most common approach is to use the volumetric representation and apply deep networks dedicated to image segmentation. However, this approach has several limitations and does not scale well into high-resolution volumes, nor takes into account the data sparsity. In our work, we reformulate the problem into a point cloud completion task. We propose an iterative, transformer-based method to reconstruct the cranial defect at any resolution while also being fast and resource-efficient during training and inference. We compare the proposed methods to the state-of-the-art volumetric approaches and show superior performance in terms of GPU memory consumption while maintaining high-quality of the reconstructed defects.

High-Resolution Cranial Defect Reconstruction by Iterative, Low-Resolution, Point Cloud Completion Transformers

TL;DR

This work reframes cranial defect reconstruction as a large-scale point-cloud completion problem and introduces an iterative, geometry-aware transformer to complete missing skull regions directly in 3-D. By replacing volumetric approaches with a sparse point-cloud representation and a DACD-based loss, the method achieves competitive reconstruction quality while dramatically reducing GPU memory usage, enabling high-resolution completion at practical speeds. The approach is validated on SkullBreak and SkullFix datasets, showing strong generalization and robustness to unseen cases, with ablations illustrating the benefits of iterative refinement and density-aware training. The findings suggest substantial practical impact for automating personalized cranial implants, reducing computational burden, and facilitating downstream 3-D printing workflows.

Abstract

Each year thousands of people suffer from various types of cranial injuries and require personalized implants whose manual design is expensive and time-consuming. Therefore, an automatic, dedicated system to increase the availability of personalized cranial reconstruction is highly desirable. The problem of the automatic cranial defect reconstruction can be formulated as the shape completion task and solved using dedicated deep networks. Currently, the most common approach is to use the volumetric representation and apply deep networks dedicated to image segmentation. However, this approach has several limitations and does not scale well into high-resolution volumes, nor takes into account the data sparsity. In our work, we reformulate the problem into a point cloud completion task. We propose an iterative, transformer-based method to reconstruct the cranial defect at any resolution while also being fast and resource-efficient during training and inference. We compare the proposed methods to the state-of-the-art volumetric approaches and show superior performance in terms of GPU memory consumption while maintaining high-quality of the reconstructed defects.
Paper Structure (11 sections, 1 equation, 2 figures, 2 tables)

This paper contains 11 sections, 1 equation, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Overview
  4. Preprocessing
  5. Network Architecture - Point Cloud Completion Transformer
  6. Objective Function
  7. Iterative Completion
  8. Postprocessing
  9. Dataset and Experimental Setup
  10. Results
  11. Discussion

Figures (2)

  • Figure 1: Visualization of the processing pipeline.
  • Figure 2: Exemplary visualization of the reconstructed point clouds / volumes for a case from the SkullBreak dataset. The PCs are shown for the defect only (reconstructed vs ground-truth) for the presentation clarity.