Enhanced Alzheimer’s Diagnosis Using a Novel EfficientNetB0 Architecture with Dynamic Scale Fusion and Spatial Attention

Alzheimer's disease (AD) is a neurological disorder characterised by progressive cognitive failure and loss of memory. It causes severe disability, affecting patients' daily activities, such as remembering, thinking, planning, decision-making, performing tasks sequentially, and communicating. Currently, there is no cure available; therefore, researchers focus on finding a treatment and methods to detect and diagnose AD earlier. However, traditional approaches, including clinical assessment and MRI scanning, cannot reliably recognize the disease early. Although some Machine Learning (ML) algorithms have been applied to analyze MRI images to detect AD, several of them exhibit a high risk of overfitting and poor consideration of the spatial correlation of neuroimaging. To improve the ability to capture subtle signals of Alzheimer's development, the propose a new method named Dynamic Multi-Scale Feature Fusion-Attention-Guided Region Enhancement (DMSFF-AGRE). It utilizes a neural network called EfficientNetB0 with an innovative design of dynamic scale fusion and spatial attention techniques. Materials and Methods Research uses multi-centre MRI data collected by the Alzheimer’s disease Neuroimaging Initiative (ADNI). Conduct a comprehensive evaluation of the diagnostic performance of the designed algorithm using exactness, value of precision, probability of detection, and F1 measure metrics. The research findings reveal that our model significantly outperforms other traditional and state-of-the-art methods.