A Fully Automatic Vertebral Segmentation Method of EOS X-ray Images in Adolescent Idiopathic Scoliosis (AIS)
Other conference paper


Full Text

Other information
AbstractIntroduction: Three-dimensional (3D) reconstructions of the spine play an important role in the diagnosis and orthopaedic treatment of adolescent idiopathic scoliosis (AIS). The EOS system is capable of simultaneous capturing of frontal and lateral X-ray images, using ultra-low radiation doses. The simultaneous acquisition of 2 two-dimensional orthogonal images makes it possible to obtain a 3D reconstruction of the spine, but the manual calibration is laborious and operatordependent. This study aimed to develop an automatic vertebral segmentation method to aid the calibration process and serve as a basis for an automatic 3D reconstruction system.

Methods: Standard or micro low-dose X-ray images were acquired from patients with AIS via an EOS X-ray machine.
A total of 30 frontal images were annotated by 3 experienced operators and validated by a radiologist. Segmentation of thoracic vertebrae and lumbar vertebrae in frontal images were achieved by a deep-learning method.
Results: Our segmentation results were evaluated on 5 frontal EOS images. Preliminary results showed that the mean accuracy, intersection over union, sensitivity and specificity of our algorithm were 0.98, 0.85, 0.92 and 0.98, respectively. The total runtime of our algorithm on 1 image was about 1 second.

Discussion and Conclusion: EOS X-ray is a safe and high-quality imaging modality, which is useful for AIS screening, treatment, and long-term follow-up. An automatic vertebral segmentation approach can facilitate the manual calibration in 3D modelling and is a prerequisite for the automation of 3D spinal reconstructions, thus might help to reduce radiologists’ workload in AIS treatment.
All Author(s) ListJ. Li, R. Zhang, D. Wang, W. C. W. Chu, J. C. Y. Cheng and L. Shi
Name of ConferenceThe 36th Annual Congress of the Hong Kong Orthopaedic Association (HKOA)
Start Date of Conference05/11/2016
End Date of Conference06/11/2016
Place of ConferenceHong Kong
Country/Region of ConferenceHong Kong
Year2016
LanguagesEnglish-United Kingdom

Last updated on 2018-22-01 at 09:16