Computer Vision News - February 2017

Every month, Computer Vision News reviews a challenge related to our field. If you do not take part in challenges, but are interested to know the new methods proposed by the scientific community to solve them, this section is for you. The challenge of this month is just starting and we have interviewed Patrick Christ, a PhD student at the Technical University of Munich working in the Image Based Biomedical Modeling Group of Professor Bjoern Menze: Patrick organizes LiTS, the Liver Tumor Segmentation Challenge , around ISBI2017 which will be held in April. The website of the challenge, which is still looking for sponsors, is here . Patrick, can you tell us about the challenge that you are organizing for ISBI 2017 ? In my own PhD, I work on automatic liver and liver lesion segmentation. I have worked a lot with public data sets and clinical data sets. Unfortunately, there are only limited public data sets. In 2008, there was a challenge about liver tumor segmentation, but this data is not available anymore. Therefore, we decided to host and organize a new challenge and to collaborate with people around the world to make a very diverse challenge. We currently have six different collaborators who have donated data to this challenge. The goal of this challenge is automatic liver lesion segmentation. The clinical motivation behind this is that the liver is somehow the first organ that gets affected by primary and secondary metastasis. It is therefore used as a stage in the clinical routine. Let’s say a patient has breast cancer or lung cancer. Then the radiologists looks at the liver to check the state of the disease and assess whether a treatment response will increase or decrease lesion mass in the organ. What is the state of the art today? What the radiologists do is that they follow the so-called RECIST protocol simulator . In RECIST, the idea is to look into the liver and measure the three biggest lesions: not a volumetric assessment, but the largest diameter. If the patient has twenty lesions, only the three biggest lesions are measured. The next time it is scanned, those three lesions will be analyzed and the change will be somehow assessed. We believe that full volumetric assessment of these lesions will have much better results in describing the current state of the patient. What are the segmentation challenges that are specific to the liver? Challenge 30 Computer Vision News Challenge LiTS: Liver Tumor Segmentation Challenge (ISBI17) Organizer: Technical University of Munich Data Contributors: • Ludwig Maximilian University of Munich (Germany) • Radboud UMC (the Netherlands) • Polytechnique Montr é al & CHUM Research Center (Canada) • Tel Aviv University & Sheba Medical Center (Israel) • IRCAD (France) • The Hebrew University of Jerusalem & Hadassah University Medical Center (Israel)