Performance Evaluation of Deep Learning-Based Models for Classification of Levels of Dementia Disease Using MRI Dataset

Elderly people experience dementia which further worsens to Alzheimer’s Disease (AD). It is a gradual illness as well as incurable also. Therefore, early prediction and treatment of AD will prevent brain cell damage. Brain MR image study is a popular way to cultivate the growth of dementia and many researchers are following this path. Our study is based on brain MRI dataset which includes four classes—Non-demented, Very Mild Demented, Mild Demented, and Moderate Demented. Eighty percent of dataset is used as training dataset and 20% is used as testing dataset. Different deep learning classifiers, i.e., ResNet 50, Xception, NASNet-Mobile, and VGG-19 are used. Among different deep learning classifiers, ResNet 50 gives the highest accuracy that is Non-demented (95.47%), Very Mild Demented (95.78%), Mild Demented (95.54%), and Moderate Demented (96.67%). The work is done on the Google Colab Pro platform and implemented using Python programming language.