Strategies for imbalanced pattern classification for digital pathology
Gerald Schaefer · 2017
Pattern classification tasks in digital pathology, which involves the analysis of high resolution digital slides of tissue samples for medical diagnosis, are, like many other medical decision making processes, often imbalanced. This means that there are (many) more training samples of some classes available compared to others, while it is often the minority class(es) that are of medical interest. In this paper, we present strategies for addressing class imbalance in pattern classification problems including the development of cost-sensitive fuzzy classifiers and the derivation of ensemble classification methods, that is classifiers that employ multiple predictors, dedicated for imbalanced classification.