Balanced-MixUp for Highly Imbalanced Medical Image Classification

Highly imbalanced datasets are ubiquitous in medical image classification problems. In such problems, it is often the case that rare classes associated to less prevalent diseases severely under-represented labeled databases, typically resulting poor performance of machine learning algorithms due overfitting process. this paper, we propose a novel mechanism for sampling training data based on popular MixUp regularization technique, which refer as Balanced-MixUp. short, Balanced-MixUp simultaneously performs regular (i.e., instance-based) and balanced class-based) data. The two sets samples then mixed-up create more distribution from neural network can effectively learn without incurring heavily under-fitting minority classes. We experiment with highly dataset retinal images (55K samples, 5 classes) long-tail gastro-intestinal video frames (10K images, 23 classes), using CNNs varying representation capabilities. Experimental results demonstrate applying outperforms other conventional schemes loss functions specifically designed deal Code released at https://github.com/agaldran/balanced_mixup .