A CNN Framework for Alzheimer’s Diagnosis Using Landmark-Based Hippocampal Region Extraction from MRI

Abstract

Alzheimer’s disease (AD) is a serious neurological disorder that affects memory and cognitive function, making early diagnosis very important. The hippocampus is one of the first regions impacted by AD and serves as a reliable biomarker in MRI-based analysis. Instead of using full MRI slices, this study focuses on selecting specific slices based on hippocampal landmarks to improve classification accuracy. The aim was to identify which MRI view provides better results for AD detection. Using the ADNI dataset, a total of 4,500 MRI slices were analyzed across three views and categories with ResNet50 and LeNet models. The results showed that selected slices significantly improved performance compared to full images. Among the views, the coronal view achieved the highest accuracy, aligning with clinical practices. Additionally, the LeNet model demonstrated strong potential for AD classification. Overall, the approach enhances machine learning accuracy and supports more effective diagnosis.