CLINICAL: Targeted Active Learning for Imbalanced Medical Image Classification

Training deep learning models on medical datasets that perform well for all classes is a challenging task. It often the case suboptimal performance obtained some due to natural class imbalance issue comes with data. An effective way tackle this problem by using targeted active learning, where we iteratively add data points belong rare classes, training However, existing methods are ineffective in targeting datasets. In work, propose Clinical (targeted aCtive Learning ImbalaNced medICal imAge cLassification) framework uses submodular mutual information functions as acquisition mine critical from classes. We apply our wide-array of imaging variety real-world scenarios - namely, binary and long-tail imbalance. show outperforms state-of-the-art acquiring diverse set