Information Technology Reference
In-Depth Information
1 Introduction
Several clinical studies in orthopaedics have used three-dimensional (3D) models of
the spine for evaluating pathologies in spinal deformities like adolescent idiopathic
scoliosis (AIS). Scoliosis affects 2
-
3 % of the population. Every year, an estimated
30,000 children are
fitted with a brace, while 38,000 patients undergo spinal fusion
surgery. The 3D reconstruction of a patient
s spine has been extremely useful in the
undertaking of several studies such as the 3D evaluation of the immediate effect of
the treatment with the Boston brace system [
36
], pre- and postoperative comparison
of spine instrumentation surgery [
37
] and the 3D progression of scoliosis [
61
].
Established volumetric modalities such as magnetic resonance imaging (MRI) are
very attractive because it is noninvasive for the patient, but it is unfortunately not
suitable for a postoperative 3D evaluation due to the ringing artifacts caused by the
surgical implants, as well as being quite expensive and time-consuming. On the other
hand, X-ray computerized tomography (CT) is a more accurate modality than MRI in
terms of 3D reconstruction of bony structures, but CT exposes the patient to unac-
ceptable doses of ionizing radiations in order to reconstruct the entire spine geometry
(all thoracic and lumbar vertebrae). More importantly, both of these modalities can
not be done in the standing position. For these above mentioned reasons, biplanar
radiography is still the imaging technique which is most frequently used for the 3D
clinical assessment of spinal deformities since it allows the acquisition of data in the
natural standing posture while exposing the patient to a low dose of radiation.
Due to the 3D nature of AIS, the 3D reconstruction of the spine geometry was
also exploited with the goal of de
'
ning better indices to characterize the third
dimension of scoliosis. Stokes et al. [
59
] introduced measures in the transverse
plane to assess the effect of derotation maneuvers in surgical procedures. Under-
standing how to classify and quantify 3D spinal deformities remains a dif
cult
challenge in scoliosis. Recently, the concept of 3D vertebra vector parameters has
allowed for better measurements compared to 2D measurement [
25
,
58
]. The
Scoliosis Research Society (SRS) has therefore recognized the need for 3D clas-
si
cation and mandated the 3D Scoliosis Committee to continue their efforts
towards developing a 3D scheme for characterizing scoliosis. Duong et al. [
15
]
proposed an unsupervised fuzzy clustering technique in order to classify the 3D
spine based on global shape descriptors, while Sangole et al. [
55
] proposed a new
means to report 3D spinal deformities based on planes of maximal curvature
(PMC). More recently, a multivariate analysis using manifold learning was able to
identify four separate groups from the same cohort of thoracic deformities [
29
].
In order to generate 3D models of the patients spine from biplanar radiographic
images, a framework (Fig.
1
) will generally require the material components, as
well as the software components which usually involves an expert in radiology to
identify speci
c anatomical landmarks on the spine. This procedure is not only
time-consuming, tedious and error-prone, but the repeatability of the procedure
cannot be assured. A few methods have attempted to automate this process by using
registration techniques which incorporated ad hoc criterions or by using statistical
