International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

U-Net-based Hippocampus Segmentation Models: Advancements and Challenges

Authors :

Laya Mahmoudi¹ Majid Abbasi² Abolfazl Kanani³

1- 1. Pardis Cancer Research Center, Pardis Cancer Institute, Shiraz, Iran. 2. Faculty of Economics and Administrative Sciences Ferdowsi University of Mashhad Mashhad, Iran. 2. Faculty of Economics and Administrative Sciences Ferdowsi University of Mashhad Mashhad, Iran 2- 1. Pardis Cancer Research Center, Pardis Cancer Institute, Shiraz, Ir an. 2. School of Mechanical Engineering, Shiraz University, Shiraz, Iran. 2. School of Mechanical Engineering, Shiraz University, Shiraz, Iran. 3- 1. Pardis Cancer Research Center, Pardis Cancer Institute, Shiraz, Iran 2. Ionizing and Non-Ionizing Radiation Protection Research Center (INIRPRC), Shiraz University of Medical Sciences, Shiraz, Iran

Keywords :

U-net architecture،hippocampus segmentation،encoder-decoder structure،U-net variants

Abstract :

Hippocampus segmentation is crucial for investigating neurological disorders such as Alzheimer’s disease, schizophrenia, and major depressive disorder. While expert manual segmentation offers high accuracy, it's time-consuming and labor-intensive nature has accelerated the development of automated approaches. U-Net architecture, an encoder-decoder framework originally introduced for image segmentation, has emerged as a powerful solution, with several variants such as 3D U-Net, U-Net++, ResUNet, and Attention U-Net achieving notable performance gains. Despite the significant improvements brought by the U-Net-based deep learning models, there is still room for development, especially in recent years. This review examines advancements in U-Net variants from 2023 to 2025, focusing on architectural enhancements and key techniques (Attention & Feature Fusion Mechanisms, Data Augmentation and Preprocessing, and Automation and Efficiency). Challenges are discussed across three different perspectives: data-related (scarcity, imbalance, low image quality), technical (overfitting, computational demands), and clinical (generalizability, integration) challenges. Despite progress, persistent issue- particularly limited dataset and clinical applicability remain, underscoring the need for continued refinement of U-Net-based hippocampus segmentation models to enhance their utility in clinical diagnostics. These insights provide a foundation for developing clinically viable, next generation hippocampus segmentation models.