MICCAI 2020 Daily - Tuesday

3 Sina Amirrajab 11 Within the openGTN project ( https:// opengtn.eu ) , Sina had two options – either to go for a cardiac application or a brain and spine application . “I thought about it and I talked to my previous supervisor because he had experience in cardiac MRI,” he recalls. “In cardiac MRI you have moving objects – the heart, which is beating, and the lung, which is moving around. I found it very challenging to try to deal with this motion and to incorporate it in some way into the simulation . That is why I decided to focus my attention on cardiac MRI.” The idea behind this work has been forming since late last year, and earlier this year the team had a paper accepted at MIDL 2020 . Sina says he is particularly proud of the time he has spent collaborating with his co-authors, you can learn from already available imaging data.” When you are a researcher in the field of medical image analysis there is always a hunger for clinical data, he tells us, but you do not always have the annotations for that data to train your model. Even if you do, the annotations vary from expert to expert . This means the ground truth is changing. The idea of openGTN is to try to simulate or synthesize a large number of images with underlying ground truth labels. It uses a model to generate these labels, called true ground truth labels, and tries to simulate or synthesize images that are as realistic as possible to the real imaging data. With these realistic images combined with ground truth labels, it can train a segmentation, registration, or other network and model to analyze real images . DAILY Tu e s d a y