Augmented Mitotic Cell Count using Field Of Interest Proposal.

摘要: Histopathological prognostication of neoplasia including most tumor grading systems are based upon a number criteria. Probably the important is mitotic figures which commonly determined as count (MC), i.e. within 10 consecutive high power fields. Often area with highest activity to be selected for MC. However, since not known in advance, an arbitrary choice this region considered one cause variability and grading. In work, we present algorithmic approach that first calculates cell map deep convolutional network. This second step used construct estimate. Lastly, select image segment representing size ten fields overall proposal expert MC determination. We evaluate using dataset 32 completely annotated whole slide images, where 22 were training network test. find correlation r=0.936