Next: Results Up: Title
Page Previous: Tool Investigation Index:
Full Text Index Contents: Conference
Page
Proposed Processing Algorithm
In this section, a data processing algorithm to achieve
the following will be introduced:
-
Semi-Automation
-
Minimal computation and algorithm complexity
-
2-D Contour extraction and registration
-
Construction of 3-D surfaces from a series of 2-D contours.
The proposed processing algorithm needs an initial contour
of a specified anatomical entity in the first or the top layer CT image.
Then the contour of the entity can be refined and extracted. In the next
layer CT image, the contour of the entity can be extracted using the contour
in the previous image as initial contour. In the following, the processing
algorithm is detailed step by step.
-
Data Preprocessing
The input data are sequential CT images of fresh cadaver of size 512x512
of 12 bits. A thresholding operation is performed on each CT image. The
threshold to extract external surface of human body is 700 and the threshold
to extract surfaces of human skull is 1150. The gray scale of a pixel is
set to zero if the original value is below the threshold. Fig.1(a)
shows a CT image (c_vm1110.fre). Fig.1(b)
and Fig.1(c) show the images after thresholding
with threshold 700 and 1150 respectively. These images show that background
noise and unwanted tissue are removed cleanly after thresholding.
-
Contour Extraction and Registration
Active contour model is first applied to extract and register contours.
It consists of three steps:
-
Contour initialization
At this stage, user is prompted to input an initial contour by clicking
mouse on the image. Fig.2(a) shows an initial
contour on Fig.1(b). The initial contour should be
close to the actual contour of the entity.
-
Contour extraction
At this stage, the contour points are driven by the energy function
and converge to the actual contour. Fig.2(b)
shows the contour after the initial contour shown in Fig.2(a) converges to the outer edge of the head.
-
Contour extraction in the subsequent image
For the subsequent CT image shown in Fig.3(a)
(c_vm1111.fre), contour of the head can be extracted using the contour
in Fig.2(b) as initial contour. The extracted
contour is shown in Fig.3(b).
Level set is also used to extract contours. It has two steps: contour initialization
and contour extraction, which are basically the same as the first two steps
of active contour models. When extracting contours in the subsequent images,
contours from the previous image are used as initial contours, but the
information in the previous image won't be used in the contour extraction
process.
-
Contour Smoothing
The extracted contour may not be smooth enough due to noise in the
images. We use curve fitting to to remove the irregular jumping points,
i.e. for each contour, a B-spline curve is fitted to the data and then
new contour is generated by sampling points from the curve. This is done
by a Surfacer program.
-
Contour Alignment
After extracting and smoothing contours from all CT images, thes contour
points are stacked together to form 3-D surfaces. CT images in the Visible
Human Dataset have different space resolutions and thickness. Each contour
is assigned a z value according to the thickness of the CT images.
The center of the images are chosen as the origin of each layer and the
x, y coordinates of contour points are converted to actual
measurement with respect to the origin. Then all contour points are combined
into one entity and exported for mesh generation.
-
Mesh generation
The point cloud generated from contour alignment are exported to Nuages.
The contours are arranged according to their z values. Nuages
will generate triangular meshes from the input contours. The resulting
surface can be viewed by Geomview.
Next: Results Up: Title
Page Previous: Tool Investigation Index:
Full Text Index Contents: Conference
Page
