Intelligent Alzheimer's diagnosis and disability assessment: robust medical imaging analysis using ensemble learning...

Neurodegenerative disorder Alzheimer's disease (AD) has progressive characteristics and leads to severe cognitive impairment that reduces life quality. Disease management along with effective intervention depends on the detailed diagnosis conducted early. The proposed framework builds an ensemble system from ResNet-50 and EfficientNet-B3 to conduct automated AD diagnostics by processing high-resolution Magnetic Resonance Imaging (MRI) images. The proposed model uses ResNet-50 to extract features coupled with EfficientNet-B3 as its robust classifier which achieves high accuracy alongside generalization performance. A large, high-quality dataset comprising 33,984 MRI images was used, ensuring diverse representation of different disease stages: the study included participants with four dementia stages organized as Mild, Moderate, Non-demented, and Very Mild Demented. The research applied several comprehensive data preprocessing methods combining normalization steps with rescaling algorithms alongside noise elimination techniques to achieve enhanced performance. Performance tests on the model required examination of accuracy along with precision and recall metrics and F1-score and ROC curve area measurements. The ensemble model delivered remarkable overall accuracy reaching 99.32% while surpassing separate deep learning architectures. The confusion matrix evaluation results showed superb classification results for Mild and Moderate stages along with non-dementia cases while maintaining minimal Wrong choices in Very Mild Demented cases. Experimental findings demonstrate the strength of deep learning algorithms to detect AD disease stages accurately. The robust and accurate performance of the proposed model indicates it has potential for use in medical environments to support radiologists in their work of early-stage AD screening and treatment development. Additional research in diverse clinical environments will strive to optimize and validate the model so it can meet real-world diagnostic requirements for medical use.