Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

TEMAS AVANZADOS EN SEÑALES E IMÁGENES BIOMÉDICAS

Máster en Ingeniería Biomédica

PRACTICAL EXERCISE:

Image reconstruction for partially acquired MR images

AIM

The aim of this assignment is to look at the problem of reconstructing MR images when we
have only acquired part of the raw data. This is called partial k-space imaging.

METHODS

The algorithm must be implemented using matlab and the ‘Image Processing Toolbox’.

The acquired data is contained in the file ‘mri.mat’

A list of useful matlab commands for this problem is provided in this document. It is not
necessary to use them, and you can use any command/function.

The exercise can be done in teams of three students.

The students must provide (in the corresponding moodle task):

• One PDF document with the resulting images and the answers to the short questions
given.
• One M file with the matlab code (with comments)
PROBLEM

One important problem in MR imaging is the duration of the exploration, which can be even
longer than half an hour. This is not only uncomfortable for the patient, who cannot move
during the acquisition, but even prevents the use of MRI for some patients, who cannot
undergo an MR study without some sedative drug.

One way to reduce the scan time is to exploit the symmetry of the k-space. Ideally, MR images
are of a real nature, at least for the most common sequences used in clinical routine. If an
image is real, its Fourier transform has Hermitian symmetry, that is

M(u,v) = M*(-u,-v)

That means that samples that are symmetrically reflected across the origin are complex
conjugates. If this is the case, we could reduce the scan time by a factor of two, by collecting
only half of the data. But the actual data are corrupted by different noises and induce errors in
the symmetry, so at least 60% of the data is usually acquired.

Partial k-space reconstruction is widely used in practice, and algorithms have been developed
to obtain the image.
PRACTICAL EXERCISE

Load the data set mri.mat into matlab:

>> load('C:\directory\mri.mat')

This file contains the raw data acquired with a scanner. The raw data (or k-space) should have
256 x 256 samples, but only 60 % of the lines have been acquired. You can check the size the
data with the commands ‘who’ and ‘size’. You can see that the data format is ‘complex double’

Now, you have to reconstruct the complete (256 x 256) k-space.

First, build a square k-space filling with zeros the missing data. If you want to test visually the
result, you can use the command:

>> imagesc(log(abs(kspace)))

You can visualize the absolute value of the complex data (command ‘abs’). The use of the
logarithm (‘log’) is only for the convenience of presentation: as in the k-space there are very
large and very small numbers, details can be seen by taking logarithms (unless there are zero
values!).

You can modify the result with the command

>> colormap(‘gray’)

To visualize the color or gray intensity assigned to each pixel value, you can use:

>> colorbar

You could even see a single line of the k-space with the command

>> plot(abs(kspace(129,:)))

Before obtaining the whole k-space, you can reconstruct the image (remember: inverse 2D
Fourier transform). To see the resulting image, use the following command (taking the
absolute value of the complex numbers, and no logarithms here as the maximum and
minimum values are not too far apart):

>> imagesc(abs(image))

Now, obtain the whole k-space using the Hermitian symmetry by filling the missing data,
reconstruct the image with the new data, and compare the new image obtained with the
previous one.

Other useful commands: ifft2, flipud, fliplr, conj, real, imag.


QUESTIONS

1. Which k-space line has the largest intensities? Why?

2. Which image is better: the one reconstructed from the zero-padding k-space, or the one
reconstructed with the k-space with symmetries? Do you see differences?

3. Another way to reduce the acquisition time is to obtain only the central lines in the k-
space. After obtaining the Hermitian k-space, let’s reconstruct the image with only the
central lines: replacing with zeros all the lines except some central lines (e.g. 30 above and
30 below the central one). Which differences (if any) do you see in the image
reconstructed with such a small number of acquired lines? Why?

You might also like