Publication Details
Cranial Defect Reconstruction Using Cascaded CNN with Alignment
Španěl Michal, doc. Ing., Ph.D. (DCGM)
Herout Adam, prof. Ing., Ph.D. (DCGM)
Alingment, Computer aided design, Convolutional neural networks, Deep learnin, Defects, Medical computing, Statistical tests
Designing a patient-specific cranial implant usually requires reconstructing the defective part of the skull using computer-aided design software, which is a tedious and time-demanding task. This lead to some recent advances in the field of automatic skull reconstruction with use of methods based on shape analysis or deep learning. The AutoImplant Challenge aims at providing a public platform for benchmarking skull reconstruction methods. The BUT submission to this challenge is based on skull alignment using landmark detection followed by a cascade of low-resolution and high-resolution reconstruction convolutional neural network. We demonstrate that the proposed method successfully reconstructs every skull in the standard test dataset and outperforms the baseline method in both overlap and distance metrics, achieving 0.920 DSC and 4.137 mm HD.
@inproceedings{BUT168486,
author="Oldřich {Kodym} and Michal {Španěl} and Adam {Herout}",
title="Cranial Defect Reconstruction Using Cascaded CNN with Alignment",
booktitle="Towards the Automatization of Cranial Implant Design in Cranioplasty",
year="2020",
series="Lecture Notes in Computer Science",
pages="56--64",
publisher="Springer Nature Switzerland AG",
address="Lima",
doi="10.1007/978-3-030-64327-0\{_}7",
isbn="978-3-030-64326-3",
url="https://link.springer.com/chapter/10.1007/978-3-030-64327-0_7"
}