A hybrid learning approach for MRI-based detection of alzheimer’s disease stages using dual CNNs and ensemble classifier

• Published in: Scientific reports Vol. 15; no. 1; pp. 25342 - 10
• Main Authors: Zolfaghari, Sepideh, Joudaki, Atra, Sarbaz, Yashar
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 14.07.2025; Nature Publishing Group; Nature Portfolio
• Subjects: 692/308; 692/699; 692/700; Accuracy; Aged; Alzheimer Disease - diagnosis; Alzheimer Disease - diagnostic imaging; Alzheimer's disease; Alzheimer’s disease staging; Biomarkers; Classification; Cognitive ability; Datasets; Decision support system; Deep learning; Disease; Ensemble learning; Female; Global health; Humanities and Social Sciences; Humans; Learning; Machine Learning; Magnetic resonance imaging; Magnetic Resonance Imaging - methods; Male; multidisciplinary; Neural networks; Neural Networks, Computer; Neurodegenerative diseases; Older people; Public health; Science; Science (multidisciplinary); Support Vector Machine; Support vector machines; Deep learning; Alzheimer’s disease staging; Magnetic resonance imaging; Decision support system; Ensemble learning
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-025-11743-y

Alzheimer’s Disease (AD) and related dementias are significant global health issues characterized by progressive cognitive decline and memory loss. Computer-aided systems can help physicians in the early and accurate detection of AD, enabling timely intervention and effective management. This study presents a combination of two parallel Convolutional Neural Networks (CNNs) and an ensemble learning method for classifying AD stages using Magnetic Resonance Imaging (MRI) data. Initially, these images were resized and augmented before being input into Network 1 and Network 2, which have different structures and layers to extract important features. These features were then fused and fed into an ensemble learning classifier containing Support Vector Machine, Random Forest, and K-Nearest Neighbors, with hyperparameters optimized by the Grid Search Cross-Validation technique. Considering distinct Network 1 and Network 2 along with ensemble learning, four classes were identified with accuracies of 95.16% and 97.97%, respectively. However, using the derived features from both networks resulted in an acceptable classification accuracy of 99.06%. These findings imply the potential of the proposed hybrid approach in the classification of AD stages. As the evaluation was conducted at the slice-level using a Kaggle dataset, additional subject-level validation and clinical testing are required to determine its real-world applicability.