The following content has been provided by the University of Erlangen-Nürnberg.

So good morning everybody. We will continue today with a very exciting topic on magnetic navigation.

Before we will discuss this particular interventional procedure, let me briefly summarize where we are currently.

This semester we talk about interventional image processing and we are all aware of the difference between diagnostic procedures and interventional procedures.

Interventional procedures basically require systems that can be used while the patient is treated.

So have this very simple picture in mind, you have bloody fingers and you have to operate the system.

That's basically what we are considering these days.

We started out with a very basic chapter where we looked at concepts for edge detection

and the general problem of detecting features and images like corners or edges or homogeneous regions.

There is one powerful operator that basically makes use of the gradient directions and the principal directions of the gradients.

That's the so-called structure tensor and the properties of the structure tensor can be studied by using the eigenvalue and eigenvector decomposition.

That is basically a principal component analysis of the gradient vectors in a local neighborhood.

That's what I explained to you and then we have used this to do shutter segmentation as a particular example.

For example, shutter segmentation where I also have shown to you that the algorithm should not only take local properties into consideration

but also global properties like in the inner part we have higher intensity values, higher heterogeneities.

Outside of the bounding box we have homogeneous regions and lower valued intensities or we also have discussed one important concept

and we put also a reference into the web where you can read the basic structure of the Huff-Transform again.

By the way, the slides should be online this afternoon, also the annotated slides.

We talked about the Huff-Transform, how to detect straight lines in images.

Last week when I was playing golf in Portland you learned – I'm joking.

There's a little truth in it as well.

But please do not tell anybody, this is confidential.

You learned about smoothing operators and image smoothing.

In particular, you have seen the bilateral filter and a modern version of it.

And you also have seen a new type of detecting point features like the sift features.

And these operators are very important.

Now we can pre-process images such that they look a little better.

This is edge preserving filtering.

That means edges are not sandpapered.

They are not blurred.

That's important because in X-ray imaging, for instance, we are required to reduce dose to a certain amount.

And dose reduction always implies lower image quality.

With these pre-processing methods we can increase the image quality.

Then we have learned operators to detect very specific points in the image.

This will become important today and the next week with sift features or with a tensor,

structure tensor based corner detectors.

So the picture is growing, right?

The picture is growing.

And Jakob Wasser hopefully also explained to you something about the GPU implementation.

Bilateral filtering is a very time consuming operation.

And this has to be mapped on a graphics card to gain the performance that is required for interventional procedures.

So very, very interesting also from a computer scientific point of view

or from the perspective of a computer scientist.

Good.

So that's the overall big picture.

And now let's look into one bloody procedure followed by the other.

The color is already chosen right.

So you know that many diseases can be treated by minimally invasive procedures.

And most of, one of the most serious diseases are diseases of the vessel system.

So there are basically two areas in the human body.