Deep Feature Representation and Multi-Kernel SVM Model for Alzheimer’s Disease Diagnosis and Dementia Stage Prediction Using Magnetic Resonance Images

Key words:  Alzheimer’s disease  dementia  deep feature  SVM  DenseNet201

Abstract:

The proposed study presents a noble hybrid model for Alzheimer’s disease progression diagnosis using DenseNet201 feature extraction and Multi-kernel Support Vector Machine(M-SVM) classification in two stages. Firstly, participants were classified into Alzheimer’s Disease (AD), Cognitive Impairment (CI) or normal group by using DenseNet201 feature extraction and M-SVM which achieved validation accuracy of 96.6% with AUC of 0.98 and test accuracy of 97.8% with AUC of 1.0. The second phase reached validation accuracy of 99.5% and an AUC of 1.0 while classifying individuals into non-dementia, mild dementia, moderate dementia, and very mild dementia groups and obtained a test accuracy of 99.8% with an AUC of 1.0. The proposed methodology demonstrates high accuracy and reliability which delivers valuable information for understanding AD progression and serves as an effective diagnostic tool for early detection and differentiation of cognitive impairments and dementia stages. Combining DenseNet201 and M-SVM demonstrates potential benefits for Alzheimer’s disease management by improving clinical assessments and treatment approaches for patient care.