Biomedical Engineering Reference
In-Depth Information
3.2.1
Data Driven Methods
The way to quantify the movement that involves the least interference with clinical
protocols and usually do not require additional hardware is the one that uses the
clinical images that are acquired. This is the primarily reason why these methods
are attractive and preferred in the clinical setting.
In the 70s, some methods for determining the movement of translation were
proposed: for liver scintigraphy, based on the centroid of the image or, in more
simplified versions, just based on the centroid of the coordinate yy of the gamma
camera [
43
-
45
]. Also for scintigraphic images, the use of two external sources
attached to the patient's body was proposed. The images of the sources allow an
easy detection of motion [
46
].
The detection for tomographic images becomes more complex. Several different
processes have been suggested which can be classified into five groups. The methods
can then be classified as: cross-correlation (CC) [
47
-
49
], diverging squares [
50
],
two-dimensional fit (2DF) [
51
], external radioactive markers (EM) [
28
,
52
]and
optical flow (OF) [
53
].
All these methods try to estimate motion from the projections which, in the case
of SPECT, are the planar images.
CC between two data vectors,
V
and
V
0
, is based on the equation:
X
m
V.j/ V
0
.j
CC.i/ D
C i/;
(19)
j
D
1
where
m
represents the number of existing values in each vector. Variable
i
is the
deviation (in pixels) to be determined and it is usual to consider
10 i 10
.
The movement is then calculated by considering the CC between vectors
obtained from consecutive images. Generally, for defining the vectors horizontal
and/or vertical profiles are used, which are chosen either from a region of the image
[
47
] or from the entire image [
48
].
The DS algorithm was developed for application to myocardial perfusion tomo-
graphic imaging. Initially, in the first projection image the user defines a rectangle
of 10 x 10 pixels inside the left ventricle. Four new rectangles of 11
11 pixels are
defined such that each square contains the initial rectangle and also one of the four
possible adjacent rows and columns. From these four rectangles, it is chosen the one
that has the highest number of counts. Following the same criteria used previously,
four new rectangles of 12
12 pixels are defined. The process goes on until reaching
a square of 20
20 pixels. The center of this square is considered the “center of the
heart” for the first projection. For the remaining projections center of the heart is
determined following the same procedure. Finally, the motion is computed taking
into account the geometry associated to the rotation of the gamma camera.
The 2DF method was also developed specifically for cardiac SPECT images. The
user defines in the 45
ı
image projection (left anterior oblique projection) a circular
region of interest containing all the activity of the myocardium. Pixels contained in
