Transcript
I'd like to thank the Brainlab organization for having this opportunity to present here our work on the validation of the Smartbrush Angio. And we did that too also to facilitate the frameless treatment of AVMS.
This is my disclosure. So from the last two decades, an invasive frame was very common to use for treatment of AVMs, but this was not only used for accurate positioning of the patients so they'd lay very still during treatments but also to register a 3D CT scan with a 2D DSA scan using the localized box as you can see here. So with the new software of Brainlab, the Angio fusion was possible between the DSA and the MRI images. And this already is earlier presented in this session.
So in this study, we compare these two methods using the evasive method as a golden standard of frame reference. We had a research agreement with Brainlab and they're offering us a special software tool, where that could automatically read the log files that were written in the software for image registration and compared to transformation matrices of both registrations for which the direct numbers of the registrations, and also for the new registration were present. So the difference was present. In total, we did 69 registrations in 52 patients unselected, and we did that with two observers. So we also want to monitor the inter-observer variability. I was one of the observers. From 16 patients had also multiple DSA series, so we had in total 40 carotid registrations and 29 vertebral registration. All patients had all other, I think, very common MRA series and CT series. And we also included patients with prior embolization or hemorrhage.
These are the different steps in the Smartbrush Angio, and you already seen that in the movie in the earlier presentation. I'd like to go into a little bit of more detail because before we tested, we were very uncommon with the software. And the first step you want to have to do is manually adjust the vessel tree to the DSA image by scaling the vessel tree and also by rotating and have to also translations. And this kind of was a learning experience so we had to learn that in I think 20 or 30 registrations first before very get really used to that because this also defines finally the automatic fusion results.
So, well it's dependent on your manual initial positioning.
So here you can see our results and also one example of a very nice registration between the vessel tree and the DSA image. The main difference between the new registration compared to the golden standard was around 1.1 millimeter and 1.3 degree, and the mean 3D vector we also calculated that it was 1.5 millimeter plus or minus 1SD. We saw no significant difference between the two observers and we had a mean inter-observer disagreement about 0.3 millimeter. And we also found no significant difference if patients were having an absence of presence of prior hemorrhage or prior embolization or between the carotid or vertebral artery. However, there can be some other notes made. I'm not going into the detail of this histogram, but what you can see here is that the majority of the registrations were within one-half millimeter, one-half degree, but there were some outlines there and there were also some high outlines there. So we looked further into detail because we didn't understand these huge numbers.
We had nine registrations that were more than three millimeter and three degrees. And when we have one patient reviewed, we repeat the registration and the manual initial positioning, and we found that after that one, that one was almost perfect within a millimeter. So actually, we found also that the manual initial positioning was also a learning curve there.
Furthermore, the other registration were mainly always caused by the quality issue of the MRA scan or the DSA. We had also included... We had an unselected patient group so we had no upfront requirements for image quality of both scans. We also included some earlier patients which had also an angiogram on an old machine. And what we saw is here is an example of one of the difficult registrations and the high outlier is that we only had a vessel tree not on the main feeding arteries and also not on the main artery. So, there was a small field of view of the MRI scans on the MRI. So we had a three-month with radiology, but by that time it didn't matter too much because we used the invasive frame, but in the software here, well there are a lot of vessels missing and this can lead to, well, not really, very accurate fusions.
Here, you can see an example of a well, that quality of one of the DSA series. So this was also hard to do, and we also had to exclude some patients for this study.
So some short notes about our clinical practice before I go to my conclusions, what we do nowadays in the new version of the software is that we always check the quality of the MRI scan. Is it long enough scan? Does it fill the view enough? And do we see enough vessels to make a very accurate fusion? And also the quality of the DSA scans, but nowadays, we have a very nice angiography machine.
What we also do and that you can see here, if I talk about between the sides, is that we adjust the vessel tree because here you see a lot of vessels overlapping actually. And we only need from the left side, we only need this part actually and it's overlapping if you don't do anything about it. But also here, and this makes it very difficult for the software to do an automatic fusion.
So what do we do now? And I know there's a new version coming at the end of this year, but we now manually adjust the vessel tree to the vessels we really want to see. And this will be automatically integrated in the new software that you can select the left or the right carotid artery to have this removed and have a better evaluation, visual evaluation of your results.
So the delineation of the nidus then, it can then be performed on the 2d DSA images. We already saw that. And it's checked with all the other MRI Series and 3D series we make. So we will also check that on the CT scan with or without contrast on the MRA scan. And we also have a special sequence MR series where we can have time slots of 150 milliseconds with six frames per second, where you can also see the filling of the nidus is on this MR scan. And of course, we heard that well, I think still the golden standard is the DSA image. It's still very useful to have all these different 3D scans as well. And then the delineation and the fuse datasets can be exported to the eclipse system and iPlan or iPlan. Usually, nowadays you still iPlan for having a treatment plan with five dynamic of formal arts, but we also did in the eclipse system, and also the fuse datasets are seen there. The treatment is done in our Institute on the Novalis Tx machine using the 6D ExacTrac image guidance and the brain help mask.
So then I come to my conclusions that the use of the Smartbrush Angio is both feasible and accurate and it facilitates, of course, also the fractionation and the frameless approach. And the largest errors were almost exclusively were due to poor quality of the scans, and we're mainly also seen in the vertebral artery. I think this is because you have less vessels there for accurate registration. So it's a little bit harder to do sometimes in the vertebral artery. And in clinical use, we adjust the vessel tree to the vessels that are visible and that appeared to improve automatic fusion results. Well, I couldn't have done this work without our whole team for the AVM, so I'd like to thank them. And I also like to thank the team on the Novalis Tx. Thank you for your attention.
This is my disclosure. So from the last two decades, an invasive frame was very common to use for treatment of AVMs, but this was not only used for accurate positioning of the patients so they'd lay very still during treatments but also to register a 3D CT scan with a 2D DSA scan using the localized box as you can see here. So with the new software of Brainlab, the Angio fusion was possible between the DSA and the MRI images. And this already is earlier presented in this session.
So in this study, we compare these two methods using the evasive method as a golden standard of frame reference. We had a research agreement with Brainlab and they're offering us a special software tool, where that could automatically read the log files that were written in the software for image registration and compared to transformation matrices of both registrations for which the direct numbers of the registrations, and also for the new registration were present. So the difference was present. In total, we did 69 registrations in 52 patients unselected, and we did that with two observers. So we also want to monitor the inter-observer variability. I was one of the observers. From 16 patients had also multiple DSA series, so we had in total 40 carotid registrations and 29 vertebral registration. All patients had all other, I think, very common MRA series and CT series. And we also included patients with prior embolization or hemorrhage.
These are the different steps in the Smartbrush Angio, and you already seen that in the movie in the earlier presentation. I'd like to go into a little bit of more detail because before we tested, we were very uncommon with the software. And the first step you want to have to do is manually adjust the vessel tree to the DSA image by scaling the vessel tree and also by rotating and have to also translations. And this kind of was a learning experience so we had to learn that in I think 20 or 30 registrations first before very get really used to that because this also defines finally the automatic fusion results.
So, well it's dependent on your manual initial positioning.
So here you can see our results and also one example of a very nice registration between the vessel tree and the DSA image. The main difference between the new registration compared to the golden standard was around 1.1 millimeter and 1.3 degree, and the mean 3D vector we also calculated that it was 1.5 millimeter plus or minus 1SD. We saw no significant difference between the two observers and we had a mean inter-observer disagreement about 0.3 millimeter. And we also found no significant difference if patients were having an absence of presence of prior hemorrhage or prior embolization or between the carotid or vertebral artery. However, there can be some other notes made. I'm not going into the detail of this histogram, but what you can see here is that the majority of the registrations were within one-half millimeter, one-half degree, but there were some outlines there and there were also some high outlines there. So we looked further into detail because we didn't understand these huge numbers.
We had nine registrations that were more than three millimeter and three degrees. And when we have one patient reviewed, we repeat the registration and the manual initial positioning, and we found that after that one, that one was almost perfect within a millimeter. So actually, we found also that the manual initial positioning was also a learning curve there.
Furthermore, the other registration were mainly always caused by the quality issue of the MRA scan or the DSA. We had also included... We had an unselected patient group so we had no upfront requirements for image quality of both scans. We also included some earlier patients which had also an angiogram on an old machine. And what we saw is here is an example of one of the difficult registrations and the high outlier is that we only had a vessel tree not on the main feeding arteries and also not on the main artery. So, there was a small field of view of the MRI scans on the MRI. So we had a three-month with radiology, but by that time it didn't matter too much because we used the invasive frame, but in the software here, well there are a lot of vessels missing and this can lead to, well, not really, very accurate fusions.
Here, you can see an example of a well, that quality of one of the DSA series. So this was also hard to do, and we also had to exclude some patients for this study.
So some short notes about our clinical practice before I go to my conclusions, what we do nowadays in the new version of the software is that we always check the quality of the MRI scan. Is it long enough scan? Does it fill the view enough? And do we see enough vessels to make a very accurate fusion? And also the quality of the DSA scans, but nowadays, we have a very nice angiography machine.
What we also do and that you can see here, if I talk about between the sides, is that we adjust the vessel tree because here you see a lot of vessels overlapping actually. And we only need from the left side, we only need this part actually and it's overlapping if you don't do anything about it. But also here, and this makes it very difficult for the software to do an automatic fusion.
So what do we do now? And I know there's a new version coming at the end of this year, but we now manually adjust the vessel tree to the vessels we really want to see. And this will be automatically integrated in the new software that you can select the left or the right carotid artery to have this removed and have a better evaluation, visual evaluation of your results.
So the delineation of the nidus then, it can then be performed on the 2d DSA images. We already saw that. And it's checked with all the other MRI Series and 3D series we make. So we will also check that on the CT scan with or without contrast on the MRA scan. And we also have a special sequence MR series where we can have time slots of 150 milliseconds with six frames per second, where you can also see the filling of the nidus is on this MR scan. And of course, we heard that well, I think still the golden standard is the DSA image. It's still very useful to have all these different 3D scans as well. And then the delineation and the fuse datasets can be exported to the eclipse system and iPlan or iPlan. Usually, nowadays you still iPlan for having a treatment plan with five dynamic of formal arts, but we also did in the eclipse system, and also the fuse datasets are seen there. The treatment is done in our Institute on the Novalis Tx machine using the 6D ExacTrac image guidance and the brain help mask.
So then I come to my conclusions that the use of the Smartbrush Angio is both feasible and accurate and it facilitates, of course, also the fractionation and the frameless approach. And the largest errors were almost exclusively were due to poor quality of the scans, and we're mainly also seen in the vertebral artery. I think this is because you have less vessels there for accurate registration. So it's a little bit harder to do sometimes in the vertebral artery. And in clinical use, we adjust the vessel tree to the vessels that are visible and that appeared to improve automatic fusion results. Well, I couldn't have done this work without our whole team for the AVM, so I'd like to thank them. And I also like to thank the team on the Novalis Tx. Thank you for your attention.