Image Registration Methods in MRI - Advance Image Processing in Magnetic Resonance Imaging - page 209

of about 1.2 mm. In each individual heartbeat (i.e., RR interval), a maximum of
three to four slices can be acquired during the 300 msec interval positioned in
diastole. The 3-D acquisition is repeated from 40 to 80 times, in order to follow
the diffusion of the contrast agent.
Consequently, the time needed for a complete acquisition can reach 40 to
80 sec or more (40 to 80 RR intervals); in many cases the entire examination
cannot be done in breath-hold state.
The quality of the myocardial perfusion can be assessed by qualitative eval-
uation of the signal intensity in the myocardium after the CM injection. In order
to perform a quantitative analysis of the myocardial perfusion, the signal intensity
changes in the acquired images must be evaluated. Therefore, intensity/time (I/T)
curves are extracted by measuring the intensity value in the region of interest in
the myocardium during oves time. Quantitative evaluation of the signal intensity
during time provides a useful clinical index. In cardiac perfusion imaging, the
maximum slope value of the I/T curve extracted from the myocardium is related
to the vitality of the cardiac tissue.
Because the acquisition protocol is made to obtain spatial alignment of all
frames, each pixel in an image frame should correspond to the pixels in the
other frames with the same geometrical coordinates. In this case, the area of
interest selection could be done on only one image in the temporal sequence,
enhancing both the reliability and the performance of the analysis. Unfortu-
nately, obtaining spatially aligned images is a difficult task in cardiac image
acquisition, in which image misalignment due to patient breathing and poor
ECG synchronization are commonly observed. Therefore, image registration is
often needed in the postprocessing phase.
The registration methods proposed in the literature cover many of the
approaches described in the present chapter: as manual registration using
anatomical markers defined by an expert operator along all images in the
temporal sequence [43,44] and as extractions of some geometrical features
(i.e., left ventricle cavity) from each frame and image registration by regis-
tering the extracted geometrical features [45,46]. Delzescaux et al. [47] pro-
posed a method based on the manual delineation of myocardium with right
and left ventricle on one frame in the sequence. An algorithm based on
template matching then performs the sequence registration. Bidaut et al. [48]
proposed a method based on the minimization of intrinsic differences between
each image and a reference image coupled to a 2-D (i.e., three parameters)
rigid-body correction. Voxel-based methods that operate directly on the image
gray values using MI as similarity metric are effective in the present problem.
In fact, the pixel values can change in dependence from the transit of the
contrast medium. Instead of pixel-value changes the statistics of gray-levels
distribution along the images remain almost the same, leading to an MI-based
registration effective in respect to other methods.
As previously described, the number of slices that can be acquired for each
volume is strongly limited by the acquisition time. Typically, only three to four
slices can be acquired during an RR interval. Acquiring more slices means