Computer Vision News - November 2022

49 AI-enabled Medical Image Analysis slice followed by a visual transformer-based sub-network to deal with feature learning between slices, leading to a joint feature representation. Then, the most relevant slices are automatically selected, which could effectively remove the uncertain slices as well as improve the performance of the SSFL model. This work was presented by Chih-Chung Hsu. In this task, there was also another winning team, team FDVTS_COVID, from Fudan University and the Shanghai AI Laboratory, who came first in the second challenge as well. To detect COVID19 on CT scans, the authors have proposed an improved version of the strong 3D Contrastive Mixup Classification network, used as the baseline method, by adapting a pre-trained video transformer backbone to introduce natural video priors to COVID-19 diagnosis. For the COVID-19 severity detection task, the authors adapted the same framework after a segmentation step to extract both the lung region and then the inner lesion region. The lesion segmentation mask serves as complementary information for the original CT slices. The team was represented by Jilan Xu. Each winning group was awarded with a monetary prize of €200. The three prizes were sponsored by DERI, GRnet and ICCS, respectively. The workshop was organised by an international committee that joined INESC TEC and IMP Diagnostics (Portugal), the National Technical University of Athens (Greece), Radboud University Medical Center (The Netherlands), Karolinska Institutet (Sweden), Google Health (USA) and the University of Lincoln and theQueen Mary University of London (UK). which proved to be a good strategy to relieve labelling effort in the digital pathology domain. The authors were awarded with monetary prizes of €500, €300 and €150, respectively, kindly sponsored by Google. The radiology/COVID19 track included sevenpresentationsofworksonexplainable data models, medical image segmentation and classification, with a particular focus on attention-based and self-supervised models. Moreover, this track hosted six papers presenting the best-performing approaches developed for the 2nd COVID19 detection competition, four papers of the best- performing approaches developed for the COVID19 severity detection competition, as well as a paper describing the baseline models given in both competitions. The first task of the competition was won by team ACVLAB, from the National Cheng Kung University, Taiwan,whichproposed an effective spatial- slice feature learning (SSFL) for COVID-19 symptom classification on CT scans. The framework included a conventional CNN to first extract the feature embedding of each CT T OF CCV