Project page

A Baseline Approach for AutoImplant

A coarse-to-fine framework for high-resolution cranial implant generation in the MICCAI 2020 Cranial Implant Design Challenge.

Jianning LiGraz University of Technology
Antonio PepeGraz University of Technology
Christina GsaxnerGraz University of Technology
Gord von CampeMedical University of Graz
Jan EggerGraz University of Technology

Multimodal Learning for Clinical Decision Support and Clinical Image-Based Procedures, Springer, 2020

Springer arXiv Code
Overview of the coarse-to-fine cranial implant generation framework
Coarse localization followed by fine implant generation for the defect region.
Abstract

Direct implant prediction in two stages.

In this study, we present a baseline approach for AutoImplant, the cranial implant design challenge, which can be formulated as a volumetric shape learning task. The defective skull, complete skull, and cranial implant are represented as binary voxel grids. Our baseline approach predicts implants directly from defective skulls. First, an encoder-decoder network learns a coarse representation from downsampled skulls and uses it to locate the bounded defect region in the original high-resolution skull. Second, another encoder-decoder network generates a fine implant from the bounded area.

0.8555Average Dice similarity score on the test set.
5.1825 mmAverage Hausdorff distance on the test set.
Presentation

Conference presentation.

Citation

Cite this work.

@inproceedings{li2020baseline,
  title = {A Baseline Approach for AutoImplant: The MICCAI 2020 Cranial Implant Design Challenge},
  author = {Li, Jianning and Pepe, Antonio and Gsaxner, Christina and von Campe, Gord and Egger, Jan},
  booktitle = {Multimodal Learning for Clinical Decision Support and Clinical Image-Based Procedures},
  pages = {75--84},
  publisher = {Springer},
  year = {2020}
}