AI and Machine Learning Enabled Early Detection of Alzheimer's Disease Using Neuroimaging Data
Аннотация
Alzheimer's disease (AD) is a progressive neurodegenerative disorder and the leading cause of dementia worldwide, affecting an estimated 55 million individuals globally. Early and accurate detection is critical for timely therapeutic intervention and to decelerate disease progression. Conventional diagnostic methods, including clinical assessments and standard radiological evaluations, often fail to detect subtle neurological changes in the pre-clinical and prodromal stages. The advent of artificial intelligence (AI) and machine learning (ML) paradigms, particularly deep learning architectures, has opened unprecedented avenues for the automated analysis of neuroimaging data, enabling early-stage detection with high precision. This research investigates the application of diverse AI and ML algorithms — encompassing convolutional neural networks (CNNs), recurrent neural networks (RNNs), transformer-based architectures, generative adversarial networks (GANs), and ensemble methods — to structural magnetic resonance imaging (sMRI), functional MRI (fMRI), diffusion tensor imaging (DTI), and positron emission tomography (PET) data for the early detection of AD. Eight optimised model formulations (F1–F8) were developed and evaluated using multimodal neuroimaging datasets from the Alzheimer's Disease Neuroimaging Initiative (ADNI), Open Access Series of Imaging Studies (OASIS-3), and the Australian Imaging, Biomarker and Lifestyle (AIBL) study. Model performance was assessed using accuracy, sensitivity, specificity, area under the receiver operating characteristic curve (AUC-ROC), and F1-score. Among all formulations, F8 (GAN + U-Net hybrid trained on combined DTI and fMRI data) achieved the highest diagnostic accuracy of 97.1%, with AUC-ROC of 0.98 and F1-score of 0.97. The transformer-based Vision Transformer (ViT) model (F7) demonstrated accuracy of 96.8% and AUC of 0.97. Statistically significant differences were observed between deep learning models and conventional classifiers (p < 0.001). These findings validate the superior capability of deep learning-based neuroimaging analysis for early AD detection, paving the way for its clinical translation and integration into diagnostic pipelines. This study underscores the transformative potential of AI-driven neuroimaging biomarkers and provides a comprehensive framework for future research in computational neurology.