A comprehensive review of deep learning algorithms for imbalance medical image processing: recent developments and future opportunities

Abstract In the medical domain, most of the medical image datasets available today are struggling from the imbalance data problem, making it difficult to detect anomalies or rare healthcare events. Most of the rare events are found in the minority class, not in the majority class. Machine learning algorithms assume that the underlying dataset is evenly distributed. However, finding such rare medical events are difficult because of the imbalance nature of the data. Class imbalance is a major concern whether it belongs to medical or any other field. In any classification problem, the provided dataset should be equally distributed whether it is a majority class or minority class. However, in the real world we cannot observe such equal distribution in any datasets. In this review paper, we’ll discuss the primary issues involved in learning from the imbalanced dataset, as well as issues such as imbalance class problem in medical domain, by examining medical image records. In addition to that we will provide the most recent methods like GAN, OCC, DCNN, Federated Learning, Transfer Learning and Attention Mechanism for handling imbalanced datasets. Apart from this we will investigate the recent development and trends that we can use to tackle this imbalance learning problem.