Continual Learning for medical image classification

Continual Learning (CL) is a novel paradigm in which the trained model is computed via a stream of data where tasks and data are only available over-time. Indeed, such approaches are able to learn new skills and knowledge without forgetting the previous ones: no access to previously encountered data and mitigate catastrophic forgetting. In this work, we propose a comparison of different CL algorithms in performing the classification of medical images. In particular, we aim to highlight the potential and ability of current methods in preventing catastrophic forgetting of the previous tasks when a new one is learned. CL-based methods have been tested for the classification of medical images showing the viability and effectiveness of these approaches.