A Novel Approach for Classification of Dementia Using Depthwise Separable Convolution Neural Network

People with dementia experience an irreversible and progressive mental illness which impacts their capacity to remember things and think clearly along with their behavior. The initial basis for psychological deterioration leads to alterations that prevent people from accomplishing their regular daily tasks. Dementia emerges as the consequence of multiple factors including hereditary elements and environmental issues and personal life factors although scientists have not discovered all its causes accurately. The medical condition shows itself through the presence of mis-formed brain proteins. The research develops an innovative method which categorizes dementia into four stages: moderate demented, very mild, mild and non - dementia. A depthwise separable convolutional neural network (CNN) generates the groupings to facilitate honest medical diagnosis. The study employs a reconstructed MobileNet model as the main operational framework to perform depthwise separable convolution-based disease classification of dementia. Keras Image Data Generator carries out preprocessing that divided images into 224 pixels X 224 pixels segments. Image preprocessing through the Keras Image Data Generator produce segmented output images which receive a 224 x 224 pixels transformation. The segmented impersonations entered the MobileNet model through its 28 layers while convolutional operations and depthwise separable convolutions use ReLU as their non-linear activation function. The training process use the Adam optimization technique along with categorical cross-entropy for accuracy evaluation of the model. The research results demonstrated MobileNet achieves 86% classification precision and 94% average F1 score performance.