Dominique Simons, MSc - Best of the Best in Pediatric Surgery 2024

OK. And on to the next presentation. So next is Dr. Dominique Simmons, also representing IPSO. She's a pediatric surgeon at Princess Maxima Center in the Netherlands, and her presentation is regarding multimodal 3D visualization of pediatric neuroblastoma, aiding surgical planning beyond anatomic information. So we'll see it next. I'm Dominique Simmons, and I'm a PhD student in the Princess Maxima Center in the Netherlands, and I'm going to talk about multimodal 3D visualization of pediatric neuroblastoma, aiding surgical planning beyond anatomical information. I have nothing to disclose. Neuroblastoma is a surgically challenging tumor to resect, and this is mainly due to the adhesion and encasement to important structures, the heterogeneity, and the therapy-induced changes. So the goal for surgeons is to resect more than 95% of the tumor tissue while sparing those anatomical structures. And to prepare the surgeons, we always make a 3D model of the patient's specific anatomy. And this, uh, in this model, we can immediately see why a complete resection is often very difficult or even impossible, maybe due to the location of the tumor. And this can lead to the surgical dilemma of whether to resect more tumor tissue with possible complications such as a bleeding or whether to leave some residual tumor tissue. And this led to the question, can we actually distinguish vital from non-vital tumor tissue since the nerveoblastoma is so heterogeneous? To answer this question, we need to add biological information to the model as we believe that uh biology can be used as a biomarker for tumor fatality. And we want to do this by adding the ADC imaging in which diffusion restriction is associated with phyto tumor tissue and will appear as a low signal on imaging and with the MIBG SPEC CT imaging in which a high uptake is associated with phyto tumor tissue. And from these images, we can see that the uptake and the diffusion restriction change over time. And with this, we also believe that the biology of the tumor might have changed and thus also the vitality of the tumor. The preoperative model that we just saw was based on the T1 weighted MRI scan. However, the ADC scan and the MRPG scan have a different 3D position, so we have to perform image registration. And to evaluate the accuracy of the registration, we use the description which measures the overlap between two structures and the target registration error, which measures the distance between two points. And after we have aligned the images, we can use ADC and the MIBG values within the tumor volume to make a risk group delineation with thresholds uh based on literature. And when we do that for every 2D slides, we can make a 3D model, uh, which shows the biology of the tumor in which a high-risk areas um are associated with the diffusion restriction and with the MIBG uptake. And here we can see an example of an MIBG risk model. So onto the results, we included 7 patients in total in a prospective manner, and we only included patients with an abdominal neoblastoma who were eligible for surgery. We achieved an accurate initial registration for all patients with a median dies of 0.81 for the ADC and the median dies of 0.77 for the MIBG and the median target registration error was 5.3 for the ADC and 4.3 for MIBG and this allowed the creation of multi-modal 3D models for every patient. And when we look at the 3D models, we had two main findings. Firstly, uh, a similar ADC and MRIG risk model which can, which we can see over here in which uh in both models, uh, the more lateral side of the tumor was marked as high risk, whereas the more medial side of the tumor was marked as low risk. And uh we also saw that the ABC and the MIBG obtained different results and here we can see that the more um cranial part of the tumor was marked as high risk for the ADC whereas the MIBG showed a completely different region uh as high risk. And this led to the question of whether these areas really predict what the literature suggests and this brings me to, uh, my future perspectives. We should evaluate the risk model with the pathology on which we are currently working and with this, we can, um, retrospectively optimize our thresholds to create two risk groups, vital versus non-vital tumor tissue, and we can, uh, then also, um, look at the accuracy of our models and we should include more patients, of course, and maybe even go multi-center. So to conclude, we were able to match ADC and MIAG images to our T1 weighted uh models and we have created a workflow uh for creating multi-modal 3D models. Uh, however, there are still some limitations as the thresholds are based on limited literature which questions the reliability of our ABC and MIBG values that we use and we should evaluate the accuracy of the models. Uh, but we are currently working on this, so hopefully next year we can, uh, bring some exciting news. And with this, I would like to thank all of my colleagues and thank you for your attention. Yeah, thanks so much for being here. This is an amazing, um, project, and thank you for the presentation. I was wondering what kind of special equipment or software do you need to kind of render these images? Is it something that is kind of standard in imaging software that people are using worldwide, or is it something that you all have created at your center? Uh, well, I've created it myself. So we do this with, um, basic image registration. There are some software tools available for this, but we use the main tool, uh, Elastics it's called, and you do this in program, uh, programming software such as Python or, um, MATLAB, uh, something like that, but there are multiple options. How long does it take to create each of these images for? Well, the main stubborn thing to do is to make the segmentation of the neuroblastoma. That's most of the time, uh, but actually the registration is about a few minutes. It's really quick. It's Amazing. And hopefully, that will get better with time as we start, uh, learning how to do that better. Uh, I, I, uh, uh, you know, it's, it's disconcerting but exciting how you show the difference between the ADC and the MIBG, you know, what it showed. So it kind of shows us that what we think is the best way to stage now will be totally different as we learn better technologies to, it kind of throws everything out the window when we say, oh, we were wrong. This is a better way to assess. This is really great. Um, there was a question from the chat about do the radiologists create or render these 3D visualizations, or is there, is it you in your lab? Like, who does the rendering? I mainly do it myself with the help of, of course, radiologists because I always check them with them. Um, you get better over time, of course, but especially in the first phases, I always check my models with the radiologist uh to make sure that I was correct. And it's really interesting what you say about how this can possibly change because now we really believe that we should, uh, resect the MRBG positive area, but this shows that we may, maybe don't know. Right, it's, you're, you're, you're shaking everything up now. Now. Well, we still have to prove it with pathology, so that's great. There's a, there's a few other questions for you, but I think we're out of time. So if you can go ahead in the chat and answer those, I, a lot of accolades to you in the chat about your work. So congratulations. Yeah, sure, thank you.