3D brain image‐based Alzheimer's disease detection techniques using fish swarm optimizer's deep convolution Siamese neural network

Abstract

AbstractAlzheimer's disease (AD), a chronic syndrome that impacts the brain, is the most prevalent form of dementia. Dementia is a brain disease that severely affects an individual's ability to perform daily activities. It starts slowly affects the brain and creates a loss of memory, language, problem‐solving and other thinking abilities. Hence, early detection is essential to avoid the severity of this illness. Neuroimaging techniques are widely recommended diagnosing approaches by medicos for early AD detection. However, detecting AD using imaging is a challenging and time‐consuming task for human expertise. Many machine learning techniques already exist in automatic AD stages detection, but these techniques are failed to handle main issues in AD detection systems such, as preserving and identifying precise biomarker regions certainty handling and; in this research, a new convolution‐based AD stages detection framework is introduced to resolve the earlier detection system's challenges and issues. The first two convolution layers contain resizing, adaptive filtering, and adaptive histogram equalization techniques to enhance the image quality, preserving biomarker features. The third layer contains the Voxel‐based Morphometry (VBM) technique to segment the exact biomarker regions of AD stages from brain MRI images. The segmented biomarker feature is extracted and selected in the fourth and fifth layers to identify exact significant biomarker features to reduce the overfitting problem during the model training. Finally, the new food source direction investigation feature of the fish swarm optimizer (FSO) is incorporated in the deep Siamese neural network (DSNN) classification phase, which reduces the uncertainty issue during model training. The efficiency is evaluated using ADNI, AIBL, and OASIS database MRI images with various accuracy metrics. The evolution results show that the new framework is obtained a higher accuracy rate of 99.89% in AD stages detection than the comparison classifiers.