Dataset Descriptor for the AutoImplant Cranial Implant Design Challenge

Abstract

This data descriptor elaborates on a dataset that can be used for the development of automatic, data-driven approaches for cranial implant design, which is a challenging task in cranioplasty. The dataset includes 210 complete skulls as well as their corresponding defective skulls and the implants, resulting in a total of 210 × 3 = 630 files in NRRD format. We split the dataset into a training set and a test set, each containing 100 and 110 completes skulls as well as the associated defective skulls and implants, respectively. The complete skulls are segmented from the public head computed tomography (CT) collection CQ500 (http://headctstudy.qure.ai/dataset), which is licensed under CC BY-NC-SA 4.0, using thresholding (Hounsfield units ≥ 150). On each complete skull, a synthetic defect, which resembles a real defect from craniotomy, is injected. In the test set, 100 skulls have similar defects to the training set, with respect to defect size, shape and position, while the last 10 skulls have distinct defects. The whole training set and the defective skulls in the test set are released to the participants of the MICCAI 2020 AutoImplant Challenge (https://autoimplant.grand-challenge.org/). The ground truth of the test set, i.e., the complete skulls and the implants are kept private by the organizers for a single blind an objective evaluation of the participant’s results.