Abstract

This work has explored the theoretical principles of Mutual Information and Entropy and their use as combined acquisition score. The work looked at mutual information and entropy combined score values produced from a deep learning model trained to segment preclinical CT scans, which supported the investigation into its use in an active learning pipeline as an information metric, to guide the selection of data from the unlabelled data pool to incorporate into the training set. The work has shown that the combined acquisition function shows promise. Further refinement and validation are necessary to fully establish its utility in active learning tasks.