Image Registration Methods in MRI - Advance Image Processing in Magnetic Resonance Imaging - page 210

increasing the acquisition time, making almost impossible the execution of the
examination in the breath-holding state. The availability of a small number of
slices implies that data are anisotropic, in the sense that the distance between two
slices acquired in the z direction is large in respect to the in-plane resolution on
the slice. In the 3-D registration approach, this implies that when an interpolation
operation is performed to calculate the transformed volume, the interpolation in
the z direction leads to large errors. For this reason, the registration problem is
often reduced to the alignment of N 2-D images using a rigid transformation.
In this case, the number of parameters to estimate reduces to 3( N 1).

In Section 7.5, the problem of global optimization of the registration
function was described in detail. In this example, we first performed perfusion
image registration by maximizing MI along time sequence frames in respect
to a reference image for each temporal sequence. The registration is performed
using the simplex optimization method. After the first step, a more accurate
registration of each frame with the previous one using the Powell method was
made. The user has to roughly identify the left ventricle, surrounding it with
a circular mask. Without the mask, the registration algorithm may try to register
structures that do not belong to the heart region. The method is consequently
semiautomatic.
The method has been tested on two kinds of image data set. The first set was
acquired from collaborative volunteers able to hold their breath and to reduce
movements during the entire examination. The second data set was acquired from
patients with suspected CAD disease scheduled for MRI examination. For each
examination, a total of 120 images was acquired, consisting of 3 short-axis slices,
each with 40 temporal frames acquired in diastolic phase. A total of five exam-
inations on volunteers and five examinations on patients were used for algorithm
effectiveness evaluation. Therefore, a total number of 30 temporal image
sequences was used.
In order to assess the effectiveness of the automatic registration procedure,
an expert user was asked to use the program with and without the use of the
automatic registration algorithm. For each spatial slice, the endocardial and epi-
cardial contours have been manually drawn. The contours were replicated along
all frames, and the user was asked to manually correct the endocardial and
epicardial borders. We used the overlapping area (OA) index as index for the
needed correction degree. The overlapping area is the common area between the
region selected in the developing image and the reference image, normalized by
the reference area.

−

Figure 7.11 shows the average value of OA index for each frame, with and

without registration, on patient images. The value of OA index on patient images
is reduced by the registration procedure and becomes comparable with the index
measured on volunteer images.
From the presented example, we can infer that the use of an automatic
registration procedure based on maximization of the mutual information seems
to be effective in order to address the requirement of fast and automatic tools for
quantitative analysis of CM-enhanced MR images.