eduzhai > Applied Sciences > Engineering >

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

  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: In this study, we present a baseline approach for AutoImplant(this https URL) - the cranial implant designchallenge, which, as suggested by the organizers, can be formulated as avolumetric shape learning task. In this task, the defective skull, the completeskull and the cranial implant are represented as binary voxel grids. Toaccomplish this task, the implant can be either reconstructed directly from thedefective skull or obtained by taking the difference between a defective skulland a complete skull. In the latter case, a complete skull has to bereconstructed given a defective skull, which defines a volumetric shapecompletion problem. Our baseline approach for this task is based on the formerformulation, i.e., a deep neural network is trained to predict the implantsdirectly from the defective skulls. The approach generates high-qualityimplants in two steps: First, an encoder-decoder network learns a coarserepresentation of the implant from down-sampled, defective skulls; The coarseimplant is only used to generate the bounding box of the defected region in theoriginal high-resolution skull. Second, another encoder-decoder network istrained to generate a fine implant from the bounded area. On the test set, theproposed approach achieves an average dice similarity score (DSC) of 0.8555 andHausdorff distance (HD) of 5.1825 mm. The code is publicly available atthis https URL.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×