So welcome to the first lecture on interventional image processing.

As I said, in winter semester we talked about methods that are applied in diagnostic medical

imaging.

That means you capture an image, you have a lot of time usually, somebody is working

on the images to present it to the physician and then the radiologist does his diagnosis.

The summer semester, and I think EFA gave a very good introduction last week already,

the summer semester we will talk about interventional procedures.

And intervention is a completely different category than diagnostics because in interventional

image processing or in interventional treatment the doctor makes use of the image information

and he needs all the processing and visualization done in more or less real time.

So you always have to keep the picture in mind.

There is a patient, there is some blood, there is some equipment, some monitors and he wants

to see or the doctor wants to see images.

And the problems we have to deal with are for instance efficiency.

We have to think about efficiency.

What can we do to make methods more efficient such that you can fulfill the real time requirements?

The second requirement is robustness.

The system has to work as robust as possible.

If something goes wrong the system should not die.

At least you show half an image instead of showing no image if something goes wrong.

What mechanisms can we use to make these things more robust?

It is something that we take into consideration within this course.

What else do we have to do in interventional image processing?

We have to think about interface problems.

You can talk to the doctor and say look I worked on this brilliant algorithm.

That happened to me actually when I was working the first days for Siemens.

I talked to the doctor and said look we have 500 different parameters but you can completely

adjust your image quality to your taste.

Here you see slider 1, here you see slider 500 and this parameter is doing this and that

parameter is doing that.

The doctor will not consider these things in detail.

He wants to have a one button solution.

He doesn't care about parameterization.

So in today's systems even in those ones used during intervention there are many degrees

of freedom but you can be sure nobody is taking advantage of the degrees of freedom.

They just use the system, press the foot switch, get the images and work on the patient.

They don't want to think about how do the methods work and how can I increase the image

quality in certain situations.

That's something that should be done fully automatically.

And that's quite challenging for us computer vision image processing people to find methods

and algorithms that work quite reliably.

Another problem is for instance if you are in a business, in a global business and your

algorithm and your system has to be set up in Malaysia for instance.

And the technician there has to set up the system and has to calibrate the system and

make things as you want to have it to capture the right images with the right image quality.

How can you make sure that the technician is doing nothing wrong and all the parameters

are set properly?

How do you define a set up procedure that tells the technician look there went something

wrong in your set up.

You have to redo this or that procedure to do the geometric calibration for instance