Survey on machine learning applied to medical image analysis

This paper presents a selective survey on recent advances in machine learning applied to medical imaging. It aims to highlight both innovations that increase the performance of the models and methods that ensure certainty, interpretability and robustness of the trained models. The paper focuses particularly on new concepts such as attention modules that allow to gather specific features considering global context. Its second main focus is given to domain adaptation methods to enhance model robustness to distribution shifts. Finally, we discuss uncertainty estimation and interpretability methods to evaluate confidence in a trained model.