Let's start.

We are now in the finals, so we have two more weeks left, and today I want to finish the

chapter on rigid image registration as I did it in the previous years, and the following

two weeks I will present to you a few research results that we have achieved in this context

in terms of rigid image registration.

And the purpose of these lectures is basically to show you what is actually going on in research.

So what is the context we are currently in?

We talk about image registration, and image registration is one of the four key chapters

in this lecture on diagnostic medical image processing.

I don't want to have this pink here.

So we talk currently about registration and fusion that subsumes the process of bringing

images together into a joint coordinate system and visualizing the data, the combined data

that's meant by image fusion.

Fusion is basically the process of finding a mapping of the images into a joint coordinate

system.

And why is this one of the four key chapters?

Well, we are basically considering now the problem, assume we have captured images at

different time frames, or we have captured images with different modalities, how can

we combine the information?

So the purpose of this is combine different images, and the combination should be done

in a way that the doctor gains some additional diagnostic value out of the registered images.

Then we talked a little bit, and this was also a chapter that was very, well, general

and not going into very recent algorithms.

We talked about 3D reconstruction.

So how can we use multiple images from one and the same modality and transform them into

the volume?

So generate higher dimensional information out of the projection.

And we have to say that the reason why computer tomography wasn't available in the 20s and

30s of the last century was not due to the fact that the methods to compute it were unknown.

The reason for that was basically that no computers were available that can do all the

time consuming computations that are required to solve these line integrals.

So it took until the beginning of the 70s of the last century to build a computer tomograph,

and that was basically also the introduction of more powerful computers that time.

So this is something that is really heavily related to the achievements and advances in

computer science.

That's something we have to see.

And before that, we looked at single images, and we looked at the way how these single

images are acquired and how artifacts come into these images due to the acquisition process

and how can we eliminate things.

So we were discussing at the beginning the problem of preprocessing and acquisition specific

preprocessing, and we talked about image undistortion.

We talked about defect pixel interpolation.

We talked about the elimination of inhomogeneities in MR images.

And the interesting thing I want to tell you is that recently I received a phone call from

the MR division and what is their problem.

They have inhomogeneities and they want to have a student working on algorithms to eliminate

the inhomogeneities for certain applications.

So basically exactly that what we did in the lecture here.

And of course further and more developed algorithms are available, and we are going to look into

this in the upcoming few months with a student.