Alzheimer’s Disease Detection from MRI Scans using Deep Hybrid Architecture with Metaheuristic Hyperparameter Tuning
Abstract
Progressive neurodegenerative Alzheimer's disease (AD) has far-reaching effects on thinking, memory, and conduct. If to want better results for our patients, to need to find problems early and accurately. In this research, a deep learning-based diagnostic model is provided that uses a multi-stage architecture including temporal convolutional neural networks (TCN), convolutional neural networks (CNN), and an eXtreme Gradient Boosting (XGBoost) classifier to produce accurate results. total classification performance is improved by applying a unique technique called Hunger Games Search (HGS) for optimal hyperparameter tuning. dataset utilized comprises 6,400 pre-processed T1-weighted MRI pictures labeled across four categories of Alzheimer's disease: Non-Dementia, Very Mild, Mild, and Moderate/Severe. images were retrieved from several repositories. In order to meet requirements of pre-trained model, images are transformed to RGB format, normalized, and shrunk to 128×128 dimensions. To extract features from MRI images, a hybrid TCN-CNN module is used, which picks up on scans' spatial and temporal patterns. XGBoost classifier receives features that have been extracted and, when fine-tuned with HGS, outperforms traditional approaches in terms of accuracy, F1-score, and AUC-ROC values. suggested model's robustness and generalizability are confirmed by evaluation across many metrics and comparison with other optimization strategies. Moreover, model's balanced class-wise performance shows that it can detect even most modest forms of dementia. Neurologists can use this integrated framework to help diagnose and track progression of amyotrophic lateral sclerosis (AD) more quickly. Possible future extensions include attention-based mechanisms, longitudinal magnetic resonance imaging (MRI) data, or multi-modal fusion with genetic and cognitive biomarkers.