Okay, so good morning everybody.

Tuesday session, we are in one of the last lectures for the semester, unfortunately,

because now image registration is a very interesting topic and I would like to tell you way more

about these things.

Anyways, I will have a short introduction on rigid image registration.

That's our last topic.

And it's also called fusion or image fusion if you read the literature.

And in diagnostic medical image processing, this is basically the third big chapter on

our way or on our journey through the world of medical image processing.

We started out with preprocessing techniques.

We started out with preprocessing techniques and what we did, we looked into the acquisition

devices, into the laws from physics and the effects from physics that are used for imaging.

And then we were thinking about what type of artifacts are implied by the sensor technology

that we are using.

So, we discussed image undistortion.

That is a problem if you use electron optics in X-ray imaging to amplify the signal.

Image undistortion.

We also talked about the effect pixel interpolation.

That's a problem that shows up if you use flat panel technology.

So, most recent detectors are flat panel detectors and they are very complicated to build.

The manufacturing process is very difficult and they have a high rejection rate or they

have the problem that there are quite a few defect pixels and these defects can be eliminated

by using the sandpaper techniques that we discussed to eliminate the defect pixels.

Then we talked about MR imaging and the problem of inhomogeneities.

And there was a very lengthy section where we looked at those images.

The basic observation was that the anatomical information is overlaid by a low frequency

intensity ramp or a low frequency intensity curve and our task was to eliminate these

intensity, low frequency intensity variations algorithmically.

And in our practical research, it also turned out in the past that this is a very, very,

very important pre-processing operation.

If you don't eliminate the bias field, then all your post-processing methods like image

segmentation procedures, image registration procedures, they all run into problems due

to this type of artifact.

So this has to be eliminated and if you go to the hospital and if you look how MR scanners

are set up and which software tools are available, there are software routines for inhomogeneity

elimination of MR images.

So that's a very important thing.

There's still some research ongoing to improve the methods and as it is quite often in practice,

there is not a method that can be chosen all the time.

They all have pros and cons and in certain cases they work and in other cases they don't

work.

And we need to have a good understanding what are the properties of these methods.

Good.

So that's the chapter on pre-processing.

The second part was on reconstruction.

So to tell you the story, we have looked at single images and improved these images.

Now these single images are improved.

Now we can capture multiple images.

The question is can we do something with multiple images?

Yes, we can do reconstruction.