Advancing Medical Image Analysis with U-Net++ Hippocampal Region Segmentation

The hippocampus is a key part of the brain. Getting the hippocampus segmented correctly helps catch neurodegenerative diseases, like Alzheimer's, early on. In this research, we focused on using the U-Net++ network to segment the hippocampus with computer help. We studied the impact of different loss functions on the model and found through testing that BCEWithLogitsLoss loss functions performs the best. In the model evaluation work, we use the Dice coefficient, Jaccard, The positive predictive value (PPV), HD95 and other quantitative evaluation indicators were fully evaluated to obtain a model with better performance in segmentation results. We worked on making the model better and more accurate.