Artificial neural networks for the recognition of vertebral landmarks in the lumbar spine

Fabio Galbusera, Tito Bassani, Francesco Costa, Marco Brayda-Bruno, Alberto Zerbi, Hans Joachim Wilke

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

The diagnosis and treatment of spinal disorders often requires the measurements of anatomical parameters on radiographic projections, which is usually performed manually. Due to the non negligible degree of intra- and inter-observer variability of these measurements, a user-independent, automated method for the characterisation of the spinal anatomy is needed. Artificial neural networks are currently used for many automated tasks in which a robust, fault-tolerant performance is needed, and may prove to be useful for this task. In this paper, a novel method based on a neural network aimed to the automatic identification of vertebral landmarks is presented. A radiographic database of lumbar sagittal radiographic projections vertebrae of adult patients suffering from various spinal disorders was created. Vertebral landmarks at the projected corners of the vertebral endplates of L3 and L4 were manually identified in all images. The annotated images were used to train and test an artificial neural network in the automatic recognition of such landmarks. The values of clinically relevant anatomical parameters (disc and vertebral heights, disc wedging) were then geometrically calculated based on the predicted landmark coordinates and compared to manual measurements. The novel method proved to be able to identify vertebral landmarks, with errors and limitations which should be taken into account. Possible future applications of neural network-based methods include the automatic extraction of clinically relevant parameters from radiographic images of the lumbar spine.

Original languageEnglish
Pages (from-to)1-6
Number of pages6
JournalComputer Methods in Biomechanics and Biomedical Engineering: Imaging and Visualization
DOIs
Publication statusAccepted/In press - Dec 2 2016

Fingerprint

Spine
Neural networks
Observer Variation
Anatomy
Databases
Therapeutics

Keywords

  • automatic recognition
  • intervertebral disc
  • landmarks
  • lumbar spine
  • Neural network

ASJC Scopus subject areas

  • Biomedical Engineering
  • Radiology Nuclear Medicine and imaging
  • Computational Mechanics
  • Computer Science Applications

Cite this

Artificial neural networks for the recognition of vertebral landmarks in the lumbar spine. / Galbusera, Fabio; Bassani, Tito; Costa, Francesco; Brayda-Bruno, Marco; Zerbi, Alberto; Wilke, Hans Joachim.

In: Computer Methods in Biomechanics and Biomedical Engineering: Imaging and Visualization, 02.12.2016, p. 1-6.

Research output: Contribution to journalArticle

@article{c228664fb9124428aca537499088a493,
title = "Artificial neural networks for the recognition of vertebral landmarks in the lumbar spine",
abstract = "The diagnosis and treatment of spinal disorders often requires the measurements of anatomical parameters on radiographic projections, which is usually performed manually. Due to the non negligible degree of intra- and inter-observer variability of these measurements, a user-independent, automated method for the characterisation of the spinal anatomy is needed. Artificial neural networks are currently used for many automated tasks in which a robust, fault-tolerant performance is needed, and may prove to be useful for this task. In this paper, a novel method based on a neural network aimed to the automatic identification of vertebral landmarks is presented. A radiographic database of lumbar sagittal radiographic projections vertebrae of adult patients suffering from various spinal disorders was created. Vertebral landmarks at the projected corners of the vertebral endplates of L3 and L4 were manually identified in all images. The annotated images were used to train and test an artificial neural network in the automatic recognition of such landmarks. The values of clinically relevant anatomical parameters (disc and vertebral heights, disc wedging) were then geometrically calculated based on the predicted landmark coordinates and compared to manual measurements. The novel method proved to be able to identify vertebral landmarks, with errors and limitations which should be taken into account. Possible future applications of neural network-based methods include the automatic extraction of clinically relevant parameters from radiographic images of the lumbar spine.",
keywords = "automatic recognition, intervertebral disc, landmarks, lumbar spine, Neural network",
author = "Fabio Galbusera and Tito Bassani and Francesco Costa and Marco Brayda-Bruno and Alberto Zerbi and Wilke, {Hans Joachim}",
year = "2016",
month = "12",
day = "2",
doi = "10.1080/21681163.2016.1261370",
language = "English",
pages = "1--6",
journal = "Computer Methods in Biomechanics and Biomedical Engineering: Imaging and Visualization",
issn = "2168-1163",
publisher = "Taylor and Francis Ltd.",

}

TY - JOUR

T1 - Artificial neural networks for the recognition of vertebral landmarks in the lumbar spine

AU - Galbusera, Fabio

AU - Bassani, Tito

AU - Costa, Francesco

AU - Brayda-Bruno, Marco

AU - Zerbi, Alberto

AU - Wilke, Hans Joachim

PY - 2016/12/2

Y1 - 2016/12/2

N2 - The diagnosis and treatment of spinal disorders often requires the measurements of anatomical parameters on radiographic projections, which is usually performed manually. Due to the non negligible degree of intra- and inter-observer variability of these measurements, a user-independent, automated method for the characterisation of the spinal anatomy is needed. Artificial neural networks are currently used for many automated tasks in which a robust, fault-tolerant performance is needed, and may prove to be useful for this task. In this paper, a novel method based on a neural network aimed to the automatic identification of vertebral landmarks is presented. A radiographic database of lumbar sagittal radiographic projections vertebrae of adult patients suffering from various spinal disorders was created. Vertebral landmarks at the projected corners of the vertebral endplates of L3 and L4 were manually identified in all images. The annotated images were used to train and test an artificial neural network in the automatic recognition of such landmarks. The values of clinically relevant anatomical parameters (disc and vertebral heights, disc wedging) were then geometrically calculated based on the predicted landmark coordinates and compared to manual measurements. The novel method proved to be able to identify vertebral landmarks, with errors and limitations which should be taken into account. Possible future applications of neural network-based methods include the automatic extraction of clinically relevant parameters from radiographic images of the lumbar spine.

AB - The diagnosis and treatment of spinal disorders often requires the measurements of anatomical parameters on radiographic projections, which is usually performed manually. Due to the non negligible degree of intra- and inter-observer variability of these measurements, a user-independent, automated method for the characterisation of the spinal anatomy is needed. Artificial neural networks are currently used for many automated tasks in which a robust, fault-tolerant performance is needed, and may prove to be useful for this task. In this paper, a novel method based on a neural network aimed to the automatic identification of vertebral landmarks is presented. A radiographic database of lumbar sagittal radiographic projections vertebrae of adult patients suffering from various spinal disorders was created. Vertebral landmarks at the projected corners of the vertebral endplates of L3 and L4 were manually identified in all images. The annotated images were used to train and test an artificial neural network in the automatic recognition of such landmarks. The values of clinically relevant anatomical parameters (disc and vertebral heights, disc wedging) were then geometrically calculated based on the predicted landmark coordinates and compared to manual measurements. The novel method proved to be able to identify vertebral landmarks, with errors and limitations which should be taken into account. Possible future applications of neural network-based methods include the automatic extraction of clinically relevant parameters from radiographic images of the lumbar spine.

KW - automatic recognition

KW - intervertebral disc

KW - landmarks

KW - lumbar spine

KW - Neural network

UR - http://www.scopus.com/inward/record.url?scp=85004024868&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85004024868&partnerID=8YFLogxK

U2 - 10.1080/21681163.2016.1261370

DO - 10.1080/21681163.2016.1261370

M3 - Article

AN - SCOPUS:85004024868

SP - 1

EP - 6

JO - Computer Methods in Biomechanics and Biomedical Engineering: Imaging and Visualization

JF - Computer Methods in Biomechanics and Biomedical Engineering: Imaging and Visualization

SN - 2168-1163

ER -