Figure 4.
Illustration of the proposed region-based selection criterion: a sliding window is used to traverse the whole slide with semi-supervised learning model to produce patch-level prediction matrix M, which is then convoluted with kernel H to transform it into that contains region-level metrics.