Why do you want to even do that?

So first my PRS slide, so I'm a member since October 2016, so that's my sixth PRS and

I missed the one in the summer last time and I've currently won, so actually the talk today

is submitted currently by TMI and the review and then further publications like the Py

Ren N framework and then the help with the learning with non-operator reduction nature paper with

Andreas and all the other co-authors, right, and then two journals with Bernhard.

So let's go to the main motivation, the hybrid imaging, so if we think about that it would

be a nice thing to start to combine our main modalities like X-ray and MRI for interventional

or diagnostically used and we can think about multiple or different benefits like reduced

treatment time or treatment specificity or we have speed up, we can use the X-ray maybe

to speed up the MR acquisition or we have dose reduction as we can switch between the both

modalities or something like that and the main question which comes up here is how to acquire

both thickness and preserve the benefits of both of the modalities and we are looking especially

in the project towards an interventional MR X-ray setting where we think about catheter

based interventions for cardiac or neuro interventions to setting stents or something like that and

there are multiple challenges coming up like the workflow is mostly with on-fluoroscopic

imaging which is not maybe very beneficial for the MR device.

Also this should be working in real time otherwise stents placement and thrombectomy would not

also not be very useful for the physician if they have to wait a few minutes and also

we have to keep in mind that we need interventional devices for that like normal catheters are

not necessarily visible in the MR while it is very easy in the X-ray so this also opens

up the point where we can fuse them maybe very nicely.

And this concludes to the point that we decided to go for an MR projection imaging.

Where does this come from?

So actually if we have a patient then we can put them into our device and then we can look

at the patient in the office Fourier transforms or we make a FFT for our patient and then

we can use the MR or the X-ray to acquire the signals of that.

So actually this means we can either acquire MR signals which are arbitrary trajectories

mostly are Cartesian sampling but also means you have tomographic slices already which

is not necessarily helpful especially as your X-ray device is only to able to acquire projection

images.

And they are then so if we would sample the stuff on a polar grid or a polar trajectory

then we would have non-distorted MR projections while the X-ray of course in the normal setting

is a point source so we have perspective distorted projections.

And this means even if we acquire MR projections we cannot overlay them directly for the interventional

setting which is not very beneficial then for the setting I motivated especially for

the casitas where you think on you want to maybe show the edges very clear to the vessel

and the casita and then with one distortion and not distortion would not fit together.

And of course you can reconstruct that back to the patient domain with different reconstruction

algorithms like the X-ray then you can use an FTP algorithm or the MR with an IAFT or

ENaFT if you have a non-Cartesian sampling right.

And we have also artifacts as you can see in the patient because he's not smiling right.

So coming to the setting what we are looking at now so we are looking into parallel projections

from the MR and also directly the cone beam projection of the X-ray and we want to have

both of them in the cone beam so in the distorted projection.

So this means we have to sample multiple parallel projections with the MR to get the same visual

impression as the X-ray and I visualized this now here you can see that this is the cone

beam projection plane and we can now think about what parallel projections do we want

to measure and if I talk later in the talk from a 5x5 re-binning I mean that we take

these orange grid sampling points so means 25 parallel projections and then to try to