Attention-Based Deep Neural Network with Rectified Adam Optimization for Alzheimer's Disease Detection Using Retinal MRI Imaging

Abstract

A chronic, and degenerative brain disease is termed as Alzheimer's Disease (AD). Now, there is no effective treatment for AD. The treatment plans and patient care have been greatly improved by the Early Detection (ED) of the AD. For ED and monitoring, Magnetic Resonance Imaging (MRI) is considered to be an effective tool. Because, MRI is non-invasive and produces high quality images. The Retinal imaging is the most popular non-invasive retinal imaging technique can be used by Deep Learning (DL) models for diagnosing AD dementia. Here, the conventional methods for AD detection experience difficulties in generalizability, interpretability, and computational efficiency. These conventional methods restrict its implications in clinical settings. But, these conventional methods for AD detection models exhibit high Accuracy (Acc). An Attention based Deep Neural Network (ADNN) with rectified Adam Optimizer (AO) using the National Institute of Mental Health and Neurosciences (NIMHANS) dataset was suggested in this study for overcoming those limitations. Image normalization is the initial method for ensuring the data consistency. Then, the Histogram of Oriented Gradients (HOG) with Wavelet Transform (WT) was utilized for robust Feature Extraction (FE). For classification, the ADNN processed the extracted features. Thus, focusing on critical areas related to AD improves the interpretability. Rectified AO optimizes the model, this will result in improving the convergence speed, and reducing overfitting. Significant robustness, interpretability, and efficiency improvements was demonstrated by the outcomes. For Real-Time (RT) application in clinical AD detection, this framework is beneficial.