Abstract
A common neurological condition that causes cognitive decline, particularly in older people, is Alzheimer’s disease (AD). In order to improve patient treatment and minimize the progression of AD, early and precise recognition is essential. Complex and expert-dependent procedures are frequently used to diagnose the clinical phases of AD, which include mildly demented, very mildly demented, moderately demented, and nondemented states. Early-stage detection might be greatly advanced by the rapid growth of deep learning techniques, especially in neuroscience research of brain imaging. In the suggested model, a feature-rich architecture is used that is tailored to improve learning representation in various stages of Alzheimer’s disease. Faster processing is made possible by its lightweight architecture, which also makes it feasible to incorporate into real-time healthcare systems. This study integrates brain imaging from magnetic resonance imaging with a process of deep learning to categorize the different phases of Alzheimer’s disease. MobileNet, DenseNet121, InceptionV3, VGG19, Xception, and CustomCNN were among the many models that underwent testing. The data set used to test the model includes 83,860 brain MRI images generated from Kaggle. At 99.10% classification accuracy, the results of the experiments show that CustomCNN performs noticeably better than the other models. With an error rate of 0.009, evaluations using metrics such as confusion matrices, precision, recall, F1 score, error rate, AUC, and average mean further validate the effectiveness of the suggested model.
