Siteng Ma, UCD

Deep active learning (AL) is widely utilised to minimise labeling costs in medical image analysis. Typically, deep learning (DL) models tend to first learn from simpler patterns and easier data. However, traditional AL methods often rely on a fixed query strategy for selecting samples, which may lead the model to overemphasise data that is challenging to classify.

This focus will result in the convergence of DL models and an increase in the amount of labeled data required to train them.

To address this issue, we propose a novel Adaptive Curriculum Query Strategy for AL (ACAL) in Medical Image Classification. During the training phase, ACAL leverages Curriculum Learning principles to initially prioritise the selection of a diverse range of samples to cover various difficulty levels, facilitating rapid model convergence.

Once the distribution of the selected samples closely matches that of the entire dataset, the query strategy shifts its focus towards difficult-to-classify data based on uncertainty. This novel approach enables the model to achieve superior performance with fewer labeled samples. This research has been accepted in the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI).