Uncertainty-Guided Semi-Supervised (UGSS) mean teacher framework for brain hemorrhage segmentation and volume quantification

Solayman Hossain Emon, Tzu-Liang Bill Tseng, Michael Pokojovy, Scott Moen, Peter McCaffrey, Eric Walser, Alex Vo, Md Fashiar Rahman

Abstract: Highlights•This work enhances robustness and generalization in segmentation using uncertainty quantification and consistency loss.•It dynamically adjusts the consistency loss, enabling the model to focus on unlabeled data as training progresses gradually.•It adopts a weighted exponential moving average (wEMA) to influence different sources of information during training.•It provides an additional control layer and customization over the conventional exponential moving average.•It demonstrates the quantification of the hemorrhage volume in 3D, further elevating diagnostic and visualization abilities.