AI-based clinical imaging annotation platforms aim to ease the annotation burden in deploying deep learning models in clinical settings. Active learning (AL), a human-in-the-loop framework, can optimise model performance under labelling constraints. However, standard, supervised AL models may overfit and bias the sample selection. To address this challenge, we present a novel platform that leverages unlabelled samples through semi-supervised learning (SSL) during training to enhance the effectiveness of active learning.
License type:
Publisher Copyright
Funding Info:
This research is supported by core funding from: I2R
Grant Reference no. : N.A.