Information Technology Reference
In-Depth Information
3 Spine Metastasis CAD System Overview
Bone metastases can appear lytic, sclerotic, or anywhere in a continuum between
these extremes. Therefore, automated detection of these lesions is a complex
problem that must be broken down into manageable components. We develop a
CAD framework that can handle both types of bone metastases. The CAD system is
a supervised machine learning framework. By supplying the CAD system with
different sets of labeled training data for lytic and sclerotic bone metastases, we can
train the system to detect the two types of bone metastases separately. The
framework has two phases: training phase and testing phase. Training phase has
four stages: spine segmentation, candidate detection, feature extraction and clas-
si
er training. The testing stage also has four stages: spine segmentation, candidate
detection, feature extraction and classi
first three stages of training and
testing phases are identical. The training phase is an of
cation. The
ne-
tuned by researchers, and the testing phase is fully automatic. Each stage will be
elaborated upon in the following sections.
fl
fine- process and can be
4 Spinal Column Segmentation and Partitioning
Spine is a bony structure with higher CT value (pixel intensity) than other tissue
types. The CT value is a gauge of X-ray beam attenuation measured by the CT
scanner, and historically was normalized into a standardized scaling set referred to
as Houns
first apply a threshold of 200 HU to mask out the
bone pixels. Then a connected component analysis is conducted on the bone mask
and the largest connected component in the center of the image is retained as the
initial spine segmentation. The bounding box of the initial segmentation is used as
the search region for the following segmentation tasks.
The spinal canal links all vertebrae into a column. On a transverse cross section,
the spinal canal appears as a low intensity oval region surrounded by high density
pedicle and lamina (Fig.
3
a). The extraction of the spinal canal is essential in order
to accurately localize the spine and form the spinal column. We apply a watershed
algorithm to detect the potential spinal canal regions, and then conduct a graph
search to locate and extract the spinal canal.
The principle of the watershed algorithm [
34
] is to transform the gradient of a
gray level image into a topographic surface. The algorithm simulates the watershed
scenario by puncturing holes at the local minimum of the intensity and
eld units (HU). We
filling the
region with water. Each region
filling with water is called a catchment basin. The
spinal canal resembles a catchment basin on a 2D cross sectional image. We
adopted the watershed algorithm implementation in ITK [
35
].
The well-known over-segmentation problem of the watershed algorithm is
alleviated by merging adjacent basins. Depth of a basin is de
ned in Eq.
1
.
