in SLAM (https://onlinelibrary.wiley.com/doi/abs/10.1002/path.5800), for any N-class prediction problem, we add "non-informative" tissue slides as an (N+1)th class and train on (N+1) classes. When we create the patient-level prediction scores during deployment, we remove all tiles which were predicted to be in the "non-informative" class. The aim is to have a "SLAM" option in Deepmed
in SLAM (https://onlinelibrary.wiley.com/doi/abs/10.1002/path.5800), for any N-class prediction problem, we add "non-informative" tissue slides as an (N+1)th class and train on (N+1) classes. When we create the patient-level prediction scores during deployment, we remove all tiles which were predicted to be in the "non-informative" class. The aim is to have a "SLAM" option in Deepmed