Exploring hippocampus segmentation on unbalanced data set using U-Net-based models
Kaiyuan Wu · Applied and Computational Engineering · 2023
This paper studies the performance of training the U-Net-based models on an unbalanced data set for hippocampus segmentation. It investigates through a series of ablation studies the effect of the weights in loss function, sampling method, model architecture, and learning rate type, and compare across different trials regarding their dice score and accuracy to identify the best strategy under class imbalance. Lastly, it displays numerical and graphical results before discussing potential implications and future directions.