HER2 Overexpression Segmentation using Machine Learning Classifiers and Quasi-random Patches for Training Data Selection