Utilizing 3D magnetic source imaging with landmark-based features and multi-classification for Alzheimer’s Disease diagnosis

• Published in: Cluster computing 2024-06, Vol.27 (3), p.2635-2651
• Main Authors: Al-Rahayfeh, Amer, Atiewi, Saleh, Almiani, Muder, Jararweh, Mohammad, Faezipour, Miad
• Format: Article
• Language: English
• Subjects: Computer Communication Networks; Computer Science; Operating Systems; Processor Architectures
• ISSN: 1386-7857; 1573-7543
• DOI: 10.1007/s10586-023-04103-w

Improvements in medical imaging have accelerated the rise of computerized healthcare. Namely, Magnetic Resource Imaging (MRI) has been shown to be a reliable method for detecting Mild Cognitive Impairment (MCI), the prenominal stage of Alzheimer’s Disease (AD) (MCI). Complex nonlinear registration and tissue segmentation are needed in order to extract features from structural MRI, which increases computation costs. We suggest the diagnosis of AD utilizing landmark-based features and multi-classification from 3D MR images to solve this issue. Preprocessing, Patch extraction, Feature learning and fusion, and Classification are the successive modules that make up our proposed work. Three processes—Noise removal, Skull stripping, and Normalization—make up the Preprocessing module. A Distributed based Adaptive Median Filter  is used to remove noise, while the Hybrid Watershed Algorithm is used to remove the skull. Particle Swarm Optimization is used to choose the most suitable landmarks for patch extraction. By increasing the effectiveness of the feature learning process, this method of patch extraction also tends to increase accuracy. A Deep Polynomial Network is used to carry out a new feature learning technique. The Genetic Algorithm is used to extract the best features from the learned features. The chosen features are then combined. The given fused features are then divided into four groups by a Support Vector Machine (SVM) classifier: AD, stable MCI (sMCI), progressive MCI (pMCI), and normal control. We put our ideas into practice utilizing the MATLAB R2017b toolkit. The proposed work outperformed the SLbL technique in terms of Accuracy, Sensitivity, Specificity, F-Score, and computation time.