Identification of cognitive brain diseases using a dual-branch siamese network on structural magnetic resonance imaging data.
Authors
Affiliations (3)
Affiliations (3)
- School of Computer Science and Software Engineering, Southwest Petroleum University, Chengdu, 610500, Sichuan, China.
- Chengdu Medical College, Chengdu, 610500, Sichuan, China.
- Department of Neurology, West China School of Medicine, Sichuan University, Sichuan University Affiliated Chengdu Second People's Hospital, Chengdu, 610017, Sichuan, China. Electronic address: [email protected].
Abstract
Early diagnosis of Alzheimer's Disease is crucial for optimizing treatment efficacy, as delayed detection often limits therapeutic outcomes. Traditional diagnostic approaches, such as cognitive assessments, PET scans, and lumbar punctures, are often invasive, costly, and less accessible. To address these limitations, we propose a Dual-Branch Siamese Network aimed at enhancing the classification accuracy of Alzheimer's Disease, Mild Cognitive Impairment, and Cognitively Normal individuals using structural MRI data. Our model integrates neuroimaging features from both Subcortical Segmentation and Cortical Parcellation, leveraging their complementary strengths to improve diagnostic precision. Experimental evaluations demonstrate that our model achieves a classification accuracy of 93% on the original dataset. To further validate the model's generalizability, we tested the trained model on a separate independent test set from the new ADNI4 database (N=191). On this independent cohort, the model achieved a robust classification accuracy of 88.48%, demonstrating its potential for real-world application. Additionally, by implementing network pruning, we reduced the model's complexity by 60% without sacrificing accuracy, thereby enhancing its feasibility for clinical use. Compared to other methods, such as convolutional neural networks and ensemble learning systems, our model demonstrates superior accuracy in multi-class classification and remains competitive in binary classification tasks. Notably, our pruned model balances accuracy with efficiency, outperforming other models in terms of computational feasibility without compromising diagnostic precision. These findings highlight the potential of our approach to facilitate early diagnosis and intervention for neurodegenerative diseases like Alzheimer's Disease.