Information Technology Reference
In-Depth Information
cues from different detectors in a way that is robust to abnormal spine geometry
caused by severe diseases. Our method is validated on a large scale of CT (189) and
MR (300) spine scans. It exhibits robust performance, especially to cases with
severe diseases and imaging artifacts.
1 Introduction
Spine is one of the major organs in the human body. It includes vertebral column
and spinal cord. A human vertebral column typically consists of 33 vertebrae. 24 of
them are articulating (7 cervical, 12 thoracic and 5 lumbar vertebrae) and 9 of them
are fused vertebrae in the sacrum and the coccyx. As spine strongly correlates to
both neural and skeletal systems, various neurological, orthopaedic and oncological
studies involve the investigations of spine anatomies. In addition, due to the strong
spatial correlations between speci
c vertebrae and their surrounding organs, spine
may also be used as a vertical reference framework to describe the locations of other
organs in the trunk, e.g., transpyloric plane. Thereby, spine becomes one of the
most frequently targeted anatomies in the interpretation of medical images.
In spine image analysis, localization and labeling of vertebrae is often the first
step, which is tedious and time consuming for manual operators. This task becomes
even more challenging for disease patients. For example, since vertebrae of a strong
scoliotic spine may not be simultaneously visible in any single coronal and sagittal
slice, a manual operators have to navigate multiple slices back and forth before
localizing and labelling all vertebrae correctly. Accordingly, an automatic spine
detection algorithm, i.e., localization and labeling of vertebrae and inter-vertebral
discs, becomes an interesting research topic. A robust spine detection algorithm will
not only bene
t various clinical applications but also paves the way to other
medical image analysis tasks, e.g., body part identi
cation and cross-modality
registration, etc.
2 Literature Review
The investigation of automatic spine detection can be traced back to the 1980s [ 1 ].
The study conducted by Chwialkowski et al. [ 1 ] aimed to detect lumbar spine discs
in 2D MR slices. Prewitt edge detectors are employed to extract morphological
information from the raw images. With the development of medical image analysis
technologies, various algorithms have been applied on vertebrae or inter-vertebrae
discs detection. In [ 2 ], Inesta et al. investigated the feasibility of identifying ver-
tebrae levels using arti
cial neuron network. Active shape model was employed by
Smyth et al. [ 3 ] to locate and measure vertebrae shapes in dual energy X-ray
absorptiometry. In order to analyze intervertebral kinematics, Bifulco et al. [ 4 ]
Search WWH ::




Custom Search