Detail publikace

Pediatric Spine Segmentation and Modeling Using Machine Learning

CHALUPA, D. MIKULKA, J. FILIPOVIČ, M. ŘÍHA, K. DOSTÁL, M.

Originální název

Pediatric Spine Segmentation and Modeling Using Machine Learning

Anglický název

Pediatric Spine Segmentation and Modeling Using Machine Learning

Jazyk

en

Originální abstrakt

Scoliosis embodies the most frequent threedimensional spinal deformity in children. Only timely treatment during the growth of the spine may significantly reduce related health problems inflicted by the deformity on adults. The results obtained via conservative therapy are problematic, and a certain degree of curvature already requires surgical treatment that at the time of writing consists of repeated spinal surgeries posing a high risk of complications. The aim is to use a spine model for computer based simulation of changes in the stress on the spine during idiopathic and syndromic deformity correction via vertebral osteotomy. A machine-learning toolbox for 3D Slicer has been developed. The toolbox has a form of an application extension. Preprocessing of the data, training and usage of the classifier is possible through a simple and modern graphical user interface. The extension is capable of performing a variety of helpful tasks such as an analysis of the impact of the size of the training vector and feature selection on classifier precision. The results suggest that the training vector size can be minimized for all of the tested classifiers. Furthermore, the random forest classifier's performance seems to be resistant to training parameter changes. Support vector machine is sensitive to training parameter changes with optimal values concentrated in a narrow feature space.

Anglický abstrakt

Scoliosis embodies the most frequent threedimensional spinal deformity in children. Only timely treatment during the growth of the spine may significantly reduce related health problems inflicted by the deformity on adults. The results obtained via conservative therapy are problematic, and a certain degree of curvature already requires surgical treatment that at the time of writing consists of repeated spinal surgeries posing a high risk of complications. The aim is to use a spine model for computer based simulation of changes in the stress on the spine during idiopathic and syndromic deformity correction via vertebral osteotomy. A machine-learning toolbox for 3D Slicer has been developed. The toolbox has a form of an application extension. Preprocessing of the data, training and usage of the classifier is possible through a simple and modern graphical user interface. The extension is capable of performing a variety of helpful tasks such as an analysis of the impact of the size of the training vector and feature selection on classifier precision. The results suggest that the training vector size can be minimized for all of the tested classifiers. Furthermore, the random forest classifier's performance seems to be resistant to training parameter changes. Support vector machine is sensitive to training parameter changes with optimal values concentrated in a narrow feature space.

Dokumenty

BibTex


@inproceedings{BUT159785,
  author="Daniel {Chalupa} and Jan {Mikulka} and Milan {Filipovič} and Kamil {Říha} and Marek {Dostál}",
  title="Pediatric Spine Segmentation and Modeling Using Machine Learning",
  annote="Scoliosis embodies the most frequent threedimensional spinal deformity in children. Only timely treatment during the growth of the spine may significantly reduce related health problems inflicted by the deformity on adults. The results obtained via conservative therapy are problematic, and a certain degree of curvature already requires surgical treatment that at the time of writing consists of repeated spinal surgeries posing a high risk of complications. The aim is to use a spine model for computer based simulation of changes in the stress on the spine during idiopathic and syndromic deformity correction via vertebral osteotomy. A machine-learning toolbox for 3D Slicer has been developed. The toolbox has a form of an application extension. Preprocessing of the data, training and usage of the classifier is possible through a simple and modern graphical user interface. The extension is capable of performing a variety of helpful tasks such as an analysis of the impact of the size of the training vector and feature selection on classifier precision. The results suggest that the training vector size can be minimized for all of the tested classifiers. Furthermore, the random forest classifier's performance seems to be resistant to training parameter changes. Support vector machine is sensitive to training parameter changes with optimal values concentrated in a narrow feature space.",
  booktitle="2019 11th International Congress on Ultra Modern Telecommunications and Control Systems and Workshops (ICUMT)",
  chapter="159785",
  doi="10.1109/ICUMT48472.2019.8970894",
  howpublished="online",
  year="2019",
  month="october",
  pages="1--5",
  type="conference paper"
}