Computer Vision News - February 2020

3 Summary RA2 DREAM Challenge 25 leave them on during X-Ray. The organizers decided not to remove those f i lms from the dataset, in order to train the algorithms on chal lenging conditions that resemble those found in the real world. Organizers hope that a new automated scoring method wi l l be generated for patient care and research. The end result wi l l be “a general izable, publ icly avai lable, automated method to generate accurate, reproducible and unbiased RA damage scores to replace the current tedious, expensive, and non-scalable method of scoring by human visual inspection.” Participants are advised to go through a walkthrough video (produced by the organizers) demonstrating the process of generating SvH scores from the perspective of a radiologist. Sub-challenges The chal lenge is divided in three sub-chal lenges, only the f irst one of which is mandatory: 1) Predict overal l RA damage from radiographic images of hands and feet. 2) Predict joint space narrowing scores from radiographic images of hands and feet. 3) Predict joint erosion scores from radiographic images of hands and feet. $50,000 wi l l be awarded in total across the 3 sub-chal lenges. It ’s a DREAM chal lenge (Dialogue on Reverse Engineering and Assessment Methods) and it operates under the DREAM Principles. Do you want to learn more about the RA2-DREAM chal lenge? This is the l ink to the chal lenge website . "One of the technical chal lenges that participants need to address is the presence of wedding rings and other jewelry" "a general izable, publ icly avai lable, automated method" UAB272 (total 7 ) UAB085 (total 179) ulna lunate radius nav mul mc1 ulna lunate radius nav mul mc1 0 0 0 0 0 0 3 5 0 3 2 0 Left Hand WRIST Erosion Comparison mc1 mul nav ulna lunate White arrows point to erosion sites White Oval points to completely collapsed joint UAB272 (total7 ) cmc5 cmc4 cmc3 capnlun radcar mna 0 0 0 0 0 0 cmc3 cmc4 cmc5 mna UAB085 (total179) cmc5 cmc4 cmc3 capnlun radcar mna 0 2 2 3 1 2