Before we start with the content, of course, first my PRS slide.

This is my tenth PRS and today I'm going to talk about the last publication that is mentioned

here which I will also present at BVM this year.

Let's play one round of find the gap.

Who of you can point out to me the gap in this picture?

Yeah thank you Mark, it's very easy.

Let's zoom out a bit.

Who of you can still find the gap in this picture?

Yeah?

Okay, perfect.

Perfect place.

Okay, let's zoom out a bit more.

Who can still find the small gap or the small crack in this picture?

Yeah, well it's somewhere there but I mean I couldn't point out to you where it is.

Okay next round.

Find the gap in this picture.

Maybe before we search for the gap I have to tell you what you're actually seeing there.

So you're seeing a slice of a CT image of the knee where we have the femoral bone on

top and the tibial bone on the bottom and we injected contrast agent here to actually

mark the gap that we're looking for.

So everything that is here white is this contrast agent and this is the gap that we're looking

for.

So we're looking for this contact region between the femoral cartilage and the tibial cartilage

which are the dark regions above and below.

And also here segmented is the tibial bone.

If we have a closer look at this you can see that even though we can find this gap it's

not really a clearly defined line but more like a region of contrast agent.

So you can see that here around this line there are still some white areas.

So already for a human it's very hard to find this gap.

We did the segmentation manually on our scan set we had.

So we went through slice by slice by slice and manually segmented this thin line that

you can see here.

Of course with manual segmentations we always have two issues.

First they are very time consuming and second you can never be sure that if I'm doing this

today and tomorrow I will be segmenting the same thing.

So last year I presented to you that we could have multiple radars segment the same volume

and then if we patch those slice by slice lines together we have surfaces and here color

coded in three colors you can see what three different people would be segmenting.

You can see that they do agree on where the cartilage should be but there are some small

differences between them.

And last year I presented that we could either average over them because we say that on average

they can find the correct surface or what we can also do to save some time is we could

have one radar segment and then smooth over the segmentation because then we're coming

very close to the consensus.

Still we need one person that is segmenting.

This takes a lot of time and also repeatability is not always given.

In literature there was one approach proposed by Miller and colleagues where they did a

semi-automatic segmentation.

But as the name says it's still semi-automatic so some manual input is required and they

state that around 40 minutes are taken to segment one volume.