mammography classification model trained from image labels only

The Cancer Registry of Norway organises a population-based breast cancer screening program, where 250 000 women participate each year. interpretation the mammograms is manual process, but deep neural networks are showing potential in mammographic screening. Most methods focus on trained from pixel-level annotations, these require expertise and time-consuming to produce. Through screenings, image level annotations however readily available. In this work we present few models Norwegian dataset: holistic model, an attention model ensemble model. We compared their performance with that pretrained based international datasets. From found our local data image-level annotation gave considerably better than external data, although annotations.