Welcome back to Beyond the Patterns.

So today I have the great pleasure to introduce Roger David Soberanis Mukul.

He is a PhD student at the Chair for Computer Aided Medical Procedures and Augmented Reality

at the Technical University of Munich.

He studied his bachelor's in computer engineering and master's in computer science at the mathematics

faculty of the Autonomous University of Yucatan, Mexico.

His work focuses on deep convolutional and graph convolutional networks for medical applications

with a particular interest in medical image segmentation.

So today I have the great pleasure to announce his presentation entitled An Uncertainty-Based

Graph Convolutional Network for Organ Segmentation Refinement.

Roger it's a great pleasure to have you here and the stage is yours.

Thank you.

Thanks a lot.

I will start the presentation.

Thanks a lot for the invitation.

As I was mentioned, I will present or send work on sequential refinement using a combination

of uncertainty analysis over the outputs of a CNN and also combining with some ideas taken

from the graph conversion networks.

After that, I would like to start mentioning a little bit about me.

As was mentioned in the beginning, I came from the Chair of Computer Aided Medical Procedures

when I am currently doing a PhD in the Technical University of Munich.

And this is a nice picture of the group at least the group of years ago because so far

for the moment we are not allowed to have these meetings but this is a picture from

two or three years ago in one of the workshops that the Chair organized.

You can see we are a numerous group.

We are split in three different, four different groups focused on augmented reality, focused

on more interdisciplinary research between medical doctors and between computer scientists.

We have also a computer vision group and finally we have a medical image analysis group with

deep learning that this one is the group where I belong.

Some of the works of my interest in generally are medical image segmentation.

Like a lot working with this particular topic but also I am interested in deep learning

for medical applications.

I have been let's say playing with CNNs and GCN architectures for refinement and also

the application of the uncertainty analysis of CNNs also for refinement that is the topic

of this talk.

Apart from that I also have working with localization problems.

This is most related to the poly problem, to the poly detection problem we have in the

bottom some examples of image taken from the end of this poly detection chronoscope in

the challenge in Mica 2015.

We have some examples of polyps in the image.

I have working with some localization, regional proposal networks for the inclusion in this

image and also in some semi-supervised learning methods also for poly detection problems.

For today I would like to talk about the approach that we proposed for doing the refinement

of the organ segmentation problem.

This is a strategy for single organ segmentation and this is a general overview that I would

like to show before going into details.

At the top we have all the components that we are using, we are taking for doing the

refinement.

We have also with all these components the main idea is to formulate the problem as a

graph learning problem over the slices of the volume or the input volume and then just