Exploring Experimental Approaches for Enhancing Alzheimer's Disease Prediction with MRI Data and Deep Learning
Year: 2025
Type: conference-paper
Abstract: Alzheimer's disease (AD) is a debilitating neurodegenerative disorder that requires an accurate diagnosis for effective treatment and management. During the past two decades, substantial research has focused on developing artificial intelligence (AI) models to automate AD prediction using datasets such as the Alzheimer's Disease Neuroimaging Initiative (ADNI). Since its launch in 2004 [1], ADNI has contributed to more than 6,000 scientific publications. Despite this progress, challenges such as data heterogeneity, limited annotations, class imbalance, and the absence of standardized resources continue to hinder reproducibility and scalability. In this study, we address these limitations by demonstrating that harmonized preprocessing of ADNI magnetic resonance images significantly improves the prediction and prognosis of AD. Our pipeline includes: (1) standardized data acquisition and preprocessing to generate three subdatasets based on MRI views: AXIAL, CORONAL, and SAGITTAL; (2) classification of these subsets using deep learning models; and (3) refinement of the most promising dataset (CORONAL) through hippocampus masking to enhance diagnostic performance. We compared several models including pre-trained transfer learning architectures: ResNet50 [2], DenseNet121 [3], InceptionV3 [4] and a newly proposed hybrid model [5] that combines InceptionV3 and DenseNet121. These were evaluated across the three anatomical views to identify the most informative MRI orientation. Using the hippocampus-masked CORONAL dataset, our hybrid model achieved a classification accuracy of 99 %, outperforming individual models. These findings show that targeted preprocessing, anatomical region focus, and hybrid deep learning models can significantly enhance AD detection and classification from MRI data.
Language: English
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W4415103863
FWCI:
Cites:
Cited by:
Related to: