Dementia Identification using a Class of CNN based Methods and Transfer Learning

Through the use of transfer learning techniques and multiple Convolutional Neural Networks (CNNs), our study offers a novel method for the early identification of dementia. Through data augmentation, we ensure robustness and enhance the model's ability to extract complex patterns from MRI data by utilizing pre-trained models such as VGG19, ResNet50, and Inception V3. We improve the identification of dementia-related patterns by utilizing well-established models, presenting promising advances in our field. In addition, our study explores the theoretical underpinnings of using MRI imaging to differentiate between different phases of dementia, offering important new perspectives on the course of the disease. We initialized the models with weights of pre-trained image net equivalents and perform dataset pre-processing, segmentation. Our results well exceeds with VGG19 topping with 91.5% accuracy followed by Inception V3 as 87.19%. ResNet50 also got an impressive score of 74.76%. The future work shows the potential to identify clearer differences to be used as early diagnostic aid for Dementia in patients who don’t have neurological symptoms with mere CNNs using transfer learning.