Complete Employment of Every Potentiality for the Optimal Detection of Alzheimer

Abstract

The main intention of this paper is to detect Alzheimer disease from the input MRI image based on deep learning methods. To effectively eliminate this Alzheimer’s disease, an analytic tool becomes mandatory that is very cost-effective, readily available, and more efficient, which senses dementia much earlier before dementia becomes Alzheimer’s. To overcome this drawback, this uses deep learning methods to detect Alzheimer disease. Knowledge of Alzheimer disease is gained through an offline process. The adaptive bilateral filter clears the noise in the input brain image in the online procedure, and then histogram equalization improves contrast. Then these images are used to find the Alzheimer-affected area. The artificial bee colony segmentation method is used to find the Alzheimer area. The gaps in Alzheimer’s area are filled with the use of the ABC segmented image’s mathematical morphology method. The textural characteristics are then utilized to detect Alzheimer's disease. After the location of Alzheimer's disease is found, the next step is to identify the severity of the disease by extracting its features. This study utilizes six extraction approaches, including the local binary pattern, histogram of gradients, SIFT, transformational wavelet features, and the Zernike moment. The BeePCNN algorithm is employed for the selection of the most excellent feature after the extraction of the features. These features are finally categorized using a deep learning method named FGPCNN. To analyze the performance of the proposed approach, this work uses real-time MRI datasets. The proposed technique provides 99.23 accuracy, the sensitivity value is 99.31, and the output value is the specific value and the error rate. The pooling layer in a convolutional neural network (CNN) is commonly used to down-sample the feature maps, reducing their spatial dimensions while preserving important features. Two types of pooling layers exist: maximum pooling and average pooling. The value of the biggest pixel in the receptive field of the filter is evaluated during max pooling. On average, the average of all values in the receptive field is evaluated. The pooling layer output is provided as an input to the following convolution layer. For big maps, CNN has extremely high computer costs. CNN trains big maps slowly. To overcome the disadvantages of the above, the Fuzzy Genetic Pulse Coupled Neural Network (FGPCNN) optimization technique is proposed.