Computer Vision News - May 2018

Eugene is presenting a fairly applied piece of work that was originally for a challenge called the Liver Tumor Segmentation Challenge - LiTS . The challenge had two parts: one was at ISBI last year and then there was a continuation at MICCAI . He tells us more: “ In the liver you can have cancer of different types. You can have primary liver cancer or you can have metastatic tumors from many different sites. Many of them collect in the liver, so it’s a very common place to find cancer. It’s useful to be able to detect and delineate the tumors that appear in the liver. These are closely linked tasks. You need to be able to detect in order to be able to delineate. This delineation is called segmentation. The challenge is about doing just that. It’s useful to have the segmentation because it gives you a sense of the volume of the tumor, which is useful for tracking treatment progression and surgery planning. ” Eugene says that the two instances of the challenge have established what the state of the art is, but the state of the art is not actually at a clinical level yet . His team found it helpful to have a picture of how different methods can be compared on the same data, and their submission achieved similar performance as the other top submissions in this method. He emphasizes that they didn’t use any external data, do anything complicated in terms of pre-processing, and did no post-processing. By keeping it sufficiently simple, if the model performs well, they can start understanding what works well. 20 Saturday ISBI DAILY Oral Presentation: Liver Lesion Segmentation Informed by Joint Liver Segmentation Eugene Vorontsov Eugene Vorontsov is a PhD student at Ecole Polytechnique at the University of Montreal, with Samuel Kadoury and Christopher Pal. “ It’s useful to have the segmentation because it gives you a sense of the volume of the tumor, which is useful for tracking treatment progression and surgery planning. ” BEST OF