DeepMitosis: Mitosis detection via deep detection, verification and segmentation networks.

•A new deep learning framework for fast and accurate mitosis detection.•The approach is the first paper that utilizes deep detection networks for the mitosis detection problem.•A deep segmentation model is used to estimate the region of mitosis when only the weak label is given. With the estimated mitotic region, we infer the bounding box labels and use them to train the deep detection model more accurately.•We design a deep verification model using residual networks that can enhance the capacity of overall system to distinguish hard negatives from detections.•We validate the proposed method on two widely used mitosis detection benchmark datasets and achieve state-of-the-art performance.