Accurate segmentation of the hippocampus from magnetic resonance imaging (MRI) data is a key requirement for detecting, diagnosing, and monitoring neurodegenerative disorders, particularly Alzheimer’s disease. Deep learning has advanced this field s...
Accurate segmentation of the hippocampus from magnetic resonance imaging (MRI) data is a key requirement for detecting, diagnosing, and monitoring neurodegenerative disorders, particularly Alzheimer’s disease. Deep learning has advanced this field substantially; however, limited-size datasets still hinder model generalization and frequently cause overfitting. The present research develops a resource-efficient and high-accuracy three-dimensional U-Net architecture specifically designed for hippocampal segmentation under constrained data conditions. The proposed pipeline integrates 3D Contrast-Limited Adaptive Histogram Equalization (CLAHE) with a Selective Coefficient-Enhanced 3D Wavelet Transform (SCE-3DWT) to enhance local contrast and suppress image noise, thereby improving feature extraction. Experiments were performed using the EADC-ADNI HarP dataset consisting of 135 hippocampal MRI volumes (input size 64 × 64 × 96 voxels). The model achieved a Dice coefficient of 0.8838 and a Jaccard index of 0.7920, outperforming several recent state-of-the-art approaches. Comparative evaluation further demonstrated low segmentation errors, with an Over- Segmentation Ratio (OSR) of 0.0594 and an Under-Segmentation Ratio (USR) of 0.0569, confirming strong robustness and generalization capability. The architecture Faizaan Fazal Khan Advisor: Prof. Goo-Rak Kwon Dept. of Information and Communication Engineering Chosun University Graduate School employs a maximum filter depth of 512 and functions efficiently without transfer learning, ensuring computational accessibility for a wider range of research and clinical settings. Future investigations will incorporate post-processing enhancements, expand dataset diversity, and evaluate higher-resolution volumetric data to further improve segmentation precision and clinical relevance. Overall, this work contributes a practical, resource-conscious framework for hippocampus segmentation and supports continued progress in medical image analysis for improved Alzheimer’s disease assessment and management.