COMPARATIVE ANALYSIS OF TRANSFER LEARNING MODELS FOR HIPPOCAMPUS CLASSIFICATION IN ALZHEIMER DISEASE PATIENTS

[featured_image]
  • Version
  • Download 12
  • File Size 454.71 KB
  • File Count 1
  • Create Date November 26, 2025
  • Last Updated November 26, 2025

COMPARATIVE ANALYSIS OF TRANSFER LEARNING MODELS FOR HIPPOCAMPUS CLASSIFICATION IN ALZHEIMER DISEASE PATIENTS

ABSTRACT:

Alzheimer's disease (AD) is the most common type of dementia that affects more than 10% of the global population, and early detection is crucial in enabling effective treatment and care measures. One of the first structural alterations in Alzheimer's patients is seen in the hippocampus, a sub cortical region of the brain important for memory and learning.Deep learning has been demonstrated to be good at classifying Alzheimer patients but there is no study that has compared the performance of deep learning models on the classification of AD using hippocampus extracted from brain magnetic resonance images. This study compares the performance of three transfer learning models such as VGG19, DenseNet201, and InceptionV3 in distinguishing non-demented and demented persons based on hippocampus features. The hippocampus was segmented from T1-weighted MRI scans, and then pre-processed with normalization, resizing, and class balancing. The dataset included 94 images, with 53 Non-Demented and 41 Demented. DenseNet201 performed better than the other models used, with accuracy of 79.65%, AUC of 88.33%, sensitivity of 81.38%, precision of 80.60% and specificity of 73.47%. VGG19 provided nicely balanced results, with an accuracy of 76.99%, AUC of 89.89%, sensitivity of 68.75%, precision of 88%, and specificity of 87.76%. InceptionV3 provided the best sensitivity of 82.81%, but a comparatively lower specificity of 61.22%, with an overall accuracy of 73.45% and an AUC of 84.22%. These findings illustrate the utility of transfer learning algorithms for reliably categorizing hippocampus regions afflicted by Alzheimer's disease, with DenseNet201 emerging as the most successful model. The findings promote the incorporation of deep learning into neuroimaging procedures for improving early diagnosis of Alzheimer's disease.

Keywords: Alzheimer disease, Magnetic Resonance Imaging, Classification, Transfer Learning, Hippocampus

SHARE