MICCAI 2016 Daily - Wednesday

already have a 4D segmentation ready. That’s usually the standard pipeline when you have images coming in 4D segmentation. When you want to push further to local tracking, you do a registration. All of this is already there, but for some unknown reason, most people would use just the first frame of the segmentation and apply the registration. But why just the first frame? I’m just going to combine everything together with no extra work. It’s just a simple code so that I don’t throw away all of this information that I had before. I just use it all. MICCAI Daily: Can you guess the unknown reasons for using only the first frame? Stéphanie: I think it’s just simpler to pick one. Every thinks that the end- diastole is the reference frame, but if you missegment a little bit, then it’s going to affect all of your results in terms of curves. Whereas, by using all of the segmentation frames, then you’re going to smooth the noise and the errors toward something much more robust. MICCAI Daily : Are you going to make clinical tests about the robustness of the model? Stéphanie: We already did tests with scar vs. non-scar tissue, looking at regional volumes. Using our coupling methods on 20 patients, we tested and proved that combining both registration and segmentation does not fail on any zones to make a difference between regional volumes of scar or regional volumes for healthy muscle. We managed for every AHA zone to see the difference between scar regional volume and healthy regional volumes. Whereas, some using the other two standard approaches failed in some zones. MICCAI Daily: What other image processing techniques helped you to prepare your code? Stéphanie: On one hand, we have an independent registration algorithm which is all sparse bayesian registration, which is great. On the other hand, we have existing software available. MICCAI Daily: What are the next steps of your work? Stéphanie: I want to apply this to tagged images to be able to have more precise local strains, circumferential or radial strains. Today we have a lot of tagged images but it’s still very unclear how to process them to a quantity of numbers that can be used every day in clinical practice. Presentation 15 MICCAI Daily: Wednesday “ there is a big discrepancy, and we should not discard all the other frames in the segmentation ”