Mitosis Detection Under Limited Annotation: A Joint Learning Approach

Abstract

Mitotic counting is a vital prognostic marker of tumor proliferation in breast cancer. Deep learning-based mitotic detection is on par with pathologists, but it requires large labeled data for training. We propose a deep classification framework for enhancing mitosis detection by leveraging class label information, via softmax loss, and spatial distribution information among samples, via distance metric learning. We also investigate strategies towards steadily providing informative samples to boost the learning. The efficacy of the proposed framework is established through evaluation on ICPR 2012 and AMIDA 2013 mitotic data. Our framework significantly improves the detection with small training data and achieves on par or superior performance compared to state-of-the-art methods for using the entire training data.

Contextual Prior Constrained Deep Networks for Mitosis Detection With Point Annotations

“…[27] principal component analysis (PCA) to approach dimensionality reduction to 650 dimensions, and then train a typical traditional classifier, the random forest (RF), to accomplish mitosis classification of the discriminant features. IBM [36] devotes to the mitosis detection with few examplar datasets and trains a deep CNN for classification on image patches with an annotated category and spatial information. These methods cannot accomplish the mitosis detection task on whole-slide images in an end-to-end way and sometimes rely on feature fusion with handcrafted features, the feature extraction and classification cannot be trained and hence would limit the performance of mitosis detection.…”