A superpixel powered autoencoder technique for detecting dementia. Issue 5 (14th December 2021)

• Record Type: Journal Article
• Authors: Bansal, Deepika; Khanna, Kavita; Chhikara, Rita; Dua, Rakesh Kumar; Malhotra, Rajeev
• Other Names: Sharma Rohit guestEditor.; Gupta Deepak guestEditor.; Maseleno Andino guestEditor.; Peng Sheng‐Lung guestEditor.; Menon Varun G. guestEditor.; Khosravi Reza guestEditor.; Jolfaei Alireza guestEditor.; Kumar Akshi guestEditor.; P Vinod guestEditor.
• Is Part Of: Expert systems. Volume 39:Issue 5(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-12-14
• Subjects: autoencoder -- dementia -- histogram of oriented gradient -- magnetic resonance imaging -- superpixel; Expert systems (Computer science); 006.33
• DOI: 10.1111/exsy.12926
• Languages: English
• ISSNs: 0266-4720

Abstract

Abstract: Dementia is a neurocognitive disorder responsible for decreasing the overall quality of life for patients. The disease has emerged as a worldwide health challenge in adults in the age group of 65 years or above. Deep learning has been successfully applied for the prediction of dementia using magnetic resonance imaging. In this paper, a superpixel‐powered autoencoder technique has been proposed using a histogram of oriented gradients for extracting the relevant features. The proposed technique is capable of predicting and classifying three categories of dementia—normal, mild cognitive impairment and dementia subjects. The viability of the proposed method is established by comparing it with the other state of art models and the popular pre‐trained networks including Squeezenet, Resnet50, Resnet18, Inceptionv3, Googlenet, VGG19 and Alexnet. The experimental results establish that the proposed model has performed significantly better than the state of art models and has outperformed the popular pre‐trained networks.