Biomedical Engineering Reference
In-Depth Information
(a)
(b)
(c)
Figure 10.
Segmentation of color slide image: (a) original color slide image; (b) propa-
gation front stops in a non-boundary area; (c) result after incorporating color and texture
features into the speed function. See attached CD for color version.
the BPV texture feature in our experiment. The distance measures are: histogram
intersection, as a measure of the block-wise histogram, and the Euclidean distance,
as a measure in the BPV item. The balancing items are:
F
A
=
−
15,
β
=5,
µ
=0
.
002,
γ
=20, and
δ
=0
.
2.
5.
SEMIAUTOMATIC SEGMENTATION AND TRACKING OF SERIAL
MEDICAL IMAGES
The ChineseVisibleHuman (CVH) datawere acquired in 2002 (for males) and
2003 (for females). It is the first volumetric data representing a complete normal
human of Asian extraction [47]. Higher resolution, distinguishable blood vessels,
data completeness, and freedom from organic lesions are the main advantages
of this dataset. It is useful in a variety of research and educational applications.
At the same time, it brings a new challenges to the research of segmentation
due to the high-resolution color details without obvious boundaries. Although the
property of high resolution in details can present difficulties in segmenting theCVH
data, it offers convenience for tracking-based serial image segmentation. When
segmenting the serial dataset, a two-step scheme based on Level Set segmentation
and tracking is proposed as follows:
1.
Segmentation Step:
This step segments the first slice of volumetric data
(e.g., a stack of slices containing the target organ(s)). In this step, the speed
function is designed for the purpose of segmentation (Section 4.4), which
incorporates the gradient, texture, and color information. The user can set
