Welcome to the Monday afternoon session. We are currently considering a very general chapter

about applications of registration methods. I mean we have seen so far the variation of

calculus and the functional optimization problem that have to be solved to build a non-rigid

image registration algorithm. And the question is what are the potential applications for these

methods and we have seen so far that difference imaging is one application, that image registration

using prior is one possible application. And today we will look into the MR intensity normalization

and tomorrow I will tell you how the non-rigid image registration methods can be combined.

MR intensity normalization. Okay I also thought that we are going to talk about

for the reconstruction. Yes we did that I remember it last Monday. We did that. So we have here you

know different bias fields and then you see here different binarization techniques where

binarization techniques where you say okay I have a threshold and all intensities below this threshold

are mapped to zero all above this threshold are mapped to 1023 and then you see that different

or one in the same dish tissue class appears completely different binary image. So our

normalization is required and then we have seen this graph of course we did that graph that where

we have the histograms and we consider the 2D histograms as images and then we compute just a

mapping between the two histograms and use this mapping that we get out of the registration process

as an intensity mapping. It tells me which intensity pair is mapped to which intensity pair

over there and the core assumption of this approach is that I expect one in the same part

of the human body to be mapped to intensities they show quite a similar distribution that's the core

idea. If I capture an MR image of the human brain of course in average and from a statistical point

of view the tissue classes should show up in each and every patient more or less the same manner and

that means that the histograms are the same and if the histograms are the same or should be the

same I can use registration methods. Here we use the two dimensional histograms with a T1 and a T2

MR sequence and just consider the histograms as images and did the image fusion and we got these

corrected images. Yeah in this example we have seen last Monday as well sorry that I couldn't

remember this. Here you see the original images as they drop out of the MR scanner here you see the

segmentation results once you apply the segmentation methods on standardized images and if you take the

original images you see that in this case the segmentation completely results in a different

completely different image than this one and also here with the brain segmentation you see

there are differences. And I also pointed out last Monday that of course you can fine-tune your

segmentation method for this particular image. Of course you can fine-tune all your segmentation

methods for the considered images but for a clinical application it is important that the

method is some kind of general that you do not have to adapt and adjust parameters to the particular

situation you are in. Okay and now let's talk about 4D reconstruction and cardiac reconstruction

basically. So if you look at the human heart and then an angiographic image of the human heart you

see the coronaries they look like that. I have no idea whether this makes sense. Yeah so they go

or they are here wrapped around the heart muscle yeah and they are on the surface of the heart

muscle. And now the heart is beating that means things are moving so the heart expands and contracts

and today these cardiac systems are able to do 30 frames per second so you to acquire 30 frames

per second. For more recent applications 100 and more frames per seconds are required but that's

something what we're industry is currently working on to build machines that do four times five times

more images a second. And we have already motivated this last week I remember this. We can now capture

these images we can capture in parallel to that the ECG signal okay. The ECG signal tells you

basically in which heart state you are. This is an indirect measurement of the heart state by just

using electrodes on the surface of the patient yeah and use the knowledge about the physiology

and the anatomy of human beings to derive the signal. So if we adjust the cardiac system properly

and we synchronize here the acquisition of images with the acquisition of the 1D ECG signal we can

do the following. We get in parallel the images and the ECG signal and we can identify certain

heart states by looking at the ECG signal and then we can pick out certain projections. We can pick

out for instance this projection and this projection and various other projections that are exactly