Prediction of Alzheimer's Disease using LeNet- CNN model with Optimal Adaptive Bilateral Filtering

• Published in: International journal of communication networks and information security Vol. 15; no. 1; pp. 52 - 58
• Main Authors: Saradhi, M V Vijaya, Rao, Pinagadi Venkateswara, Krishnan, V Gokula, Sathyamoorthy, K, Vijayaraja, V
• Format: Journal Article
• Language: English
• Published: Kohat Kohat University of Science and Technology (KUST) 01.04.2023
• Subjects: Accuracy; Algorithms; Alzheimer's disease; Brain damage; Brain research; Classification; Cognitive ability; Deep learning; Dementia; Filtration; Machine learning; Magnetic resonance imaging; Medical imaging; Neural networks; Neuroimaging; Noise reduction; Parameter identification
• ISSN: 2073-607X; 2076-0930

Alzheimer's disease is a degenerative dementia that causes progressively worsening memory loss and other cognitive and physical impairments over time. Mini-Mental State Examinations and other screening tools are helpful for early detection, but diagnostic MRI brain analysis is required. When Alzheimer's disease (AD) is detected in its earliest stages, patients may begin protective treatments before permanent brain damage has occurred. The characteristics of the lesion sites in AD affected role, as identified by MRI, exhibit great variety and are dispersed across the image space, as demonstrated in cross-sectional imaging investigations of the disease. Optimised Adaptive Bilateral filtering using a deep learning model was suggested as part of this study's approach toward this end. Denoising the pictures with the help of the suggested adaptive bilateral filter (ABF) is the first stage. The ABF improves denoising in edge, detail, and homogenous areas separately. After then, the ABF is given a weight, and the Adaptive Equilibrium Optimizer (AEO) is used to determine the best possible value for that weight. LeNet, a CNN model, is then used to complete the AD organisation. The first step in using the LeNet-5 network model to identify AD is to study the model's structure and parameters. The ADNI experimental dataset was used to verify and compare the suggested technique to other models. The experimental findings prove that the suggested method can achieve a classification accuracy of 97.43%, 98.09% specificity, 97.12% sensitivity, and 89.67% Kappa index. When compared against competing algorithms, the suggested model emerges victorious.