CVPR Daily - Wednesday

While this task is fundamentally a medical imaging challenge , its unique difficulty lies in the sheer size of histopathological images, distinguishing it from conventional computer vision tasks involving standard images. “ The task is also challenging because although we have a very large whole slide image, the cancer region is very small, ” Howard explains. “ It’s a significant data imbalance that we need to deal with. We use an image processing technique to cut these large pathological images into 256 small patches . According to existing research, very good encoders can extract the features from those patches so that we can use these features to construct our heterogeneous graph. ” By combining advanced graph-based methods and a causal-driven approach to improve cancer detection and localization in histopathology images, the team’s efforts demonstrate promising outcomes and set the stage for further advancements in the field. To learn more about Howard and Fernando’s work, visit Poster 315 this afternoon from 16:30-18:30 in the West Exhibit Hall. 16 DAILY CVPR Wednesday Poster Presentation The meta-relations specified by our heterogeneous graph data structure