Automatic Segmentation of Vertebral Body using Clustering and Geometrical Features

Arpitha, Adela; Rangarajan, Lalitha

doi:10.1109/icict43934.2018.9034456

Cited by 1 publication

(1 citation statement)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The spine life system in figure 1, can be depicted as a weight bearing hard portioned skeletal structure made up of 33 bones called vertebrae separated into 5 areas in particular cervical -7 vertebrae, thoracic -12 vertebrae, lumbar -5 vertebrae, sacrum -5 vertebrae and coccyx -4 vertebrae that are stacked over each other with ductile intervertebral plates in the middle of every vertebra. A vertebra comprises of two sections: an anterior portion that forms the VB and a posterior or the vertebral bend that comprises of spinous procedures, pedicles, vertebral lower arm and other anatomical locales [4].…”

Section: Introductionmentioning

confidence: 99%

Automatic Vertebral Body Fracture Classification through PCA based Angle Estimation

Arpitha¹,

Rangarajan²,

Raghunandan³

2018

IJCA

Self Cite

View full text Add to dashboard Cite

Our objective is to contribute a reliable platform for segmenting and classifying vertebral fractures. We present a method for the automatic segmentation of the Vertebral Bodies (VB) from 2D sagittal images of the spine followed by classifying if the VB is normal or fractured. In our work, we use Dual energy X-ray Absorptiometry (DXA) images which is a standard technique to quantify Bone Mineral Density (BMD) of vertebrae that best indicates osteoporotic vertebral fracture. A certain level of preprocessing is required as the DXA images appear noisy. After image binarization, morphological operations are applied to close the missing links and to remove small invalid regions. The Region of Interest (ROI) is portioned by extricating fine contour of vertebral endplates. The count of Connected Components (CC) indicates the presence of inferior and superior endplates of the adjacent vertebrae. A complete VB is segmented by choosing those two CCs with the largest centroid distance. Principal Component Analysis (PCA) inspired angle estimation for each CC is used to classify if the segmented VB is fractured or not. Our approach has been demonstrated to be feasible both in terms of efficiency and robustness by being tested on 527 vertebrae on DXA images at a classification accuracy of 93.93% compared to a gold standard of manual annotation by expert radiologists.

show abstract

Section: Introductionmentioning

confidence: 99%