Biomedical Engineering Reference
In-Depth Information
Table 11.2:
Categorization of atlas selection strategies by number,
type, and assignment of atlases. See Sections 11.4.1 through 11.4.4 for
details and Fig. 11.6 for a schematic overview of the different methods
Selection
No. of Atlases
Assignment of Atlas
Strategy
per Raw Image
Type of Atlas
to Raw Image
IND
single
individual
fixed
SIM
single
individual
variable
AVG
single
average
fixed
MUL
multiple
individual
fixed
together with the results for the four atlas selection strategies, are presented in
Section 11.5.
11.4.1
Segmentation with a Fixed,
Single Individual Atlas
The most straight forward strategy for selection of an atlas is to use one individ-
ual segmented image. The selection can be random, or based on heuristic criteria
such as image quality, lack of artifacts, or normality of the imaged subject. This
strategy is by far the most commonly used method for creating and using an at-
las [25]. It requires only one atlas, which greatly reduces the preparation effort
as compared to the more complex methods described below.
Out of the 20 bee brains in our population, we picked the one that was
judged to have the best image quality and least artifacts. We then used this atlas
brain to segment the remaining 19 brain images. Each of the 19 raw images was
registered non-rigidly to the microscopy image of the atlas, and labeled using
the accordingly transformed atlas label image.
11.4.2
Segmentation with the Best Atlas for an Image
Suppose that instead of a single atlas, we have several atlases that originate
from several different subjects. For each image that we are segmenting, there is
one atlas that will produce the best segmentation accuracy among all available
atlases. It is obviously desirable to use this optimum atlas, which is most likely
a different atlas for each image.
