Title: Multi-view multi-input CNN-based architecture for diagnosis of Alzheimer's disease in its prodromal stages
Authors: Mohamed Amine Zayene; Hend Basly; Fatma Ezahra Sayadi
Addresses: Laboratory of 'Networked Objects, Control, and Communication Systems (NOCCS)', Electrical Engineer Department, National School of Engineering School of Sousse, BP 264, Erriadh, 4023, Sousse, Tunisia; Faculty of Sciences of Monastir, BP 56, Monastir, 5000, Monastir, Tunisia ' Laboratory of 'Networked Objects, Control, and Communication Systems (NOCCS)', Electrical Engineer Department, National School of Engineering School of Sousse, BP 264, Erriadh, 4023, Sousse, Tunisia ' Laboratory of 'Networked Objects, Control, and Communication Systems (NOCCS)', Electrical Engineer Department, National School of Engineering School of Sousse, BP 264, Erriadh, 4023, Sousse, Tunisia
Abstract: Alzheimer's disease (AD) is a progressive neurodegenerative brain disorder, the leading cause of dementia, characterised by memory loss and cognitive decline affecting daily life. Early detection is crucial for effective treatment. 18F-FDG-PET is the most accurate clinical test for AD diagnosis, yet current methods often involve laborious data preprocessing. Thus, we propose utilising deep learning techniques, known for their effectiveness. Our study introduces a 3D convolutional neural network (3D CNN) capable of learning inter and intra-slice information simultaneously. We evaluated our method on 540 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database, including normal controls (CN), early and late mild cognitive impairment (EMCI, LMCI), and AD subjects. Results demonstrate an 85.71% accuracy in CN vs. EMCI vs. LMCI vs. AD classification on the ADNI database.
Keywords: Alzheimer's disease diagnosis; FDG-PET neuroimaging data; convolutional neural networks; CNN; multi-view; multi-input.
International Journal of Biometrics, 2024 Vol.16 No.6, pp.601 - 613
Received: 06 Aug 2023
Accepted: 01 Dec 2023
Published online: 03 Oct 2024 *