Related Experiment Video
Updated: Aug 11, 2026
DeepOmicsAE: Representing Signaling Modules in Alzheimer's Disease with Deep Learning Analysis of Proteomics, Metabolomics, and Clinical Data
Published on: December 15, 2023
Deep Cross-Branch Multi-Modal Fusion Network for early Alzheimer's diagnosis
Jiaqiang Li1, Yian Gao2, Zhenghua Guan1
1School of Biomedical Engineering, Medical School, National Regional Key Technology Engineering Laboratory for Medical Ultrasound, Guangdong Key Laboratory for Biomedical Measurements and Ultrasound Imaging, Shenzhen University, Shenzhen, 518055, Guangdong, China.
A new deep learning model, DCMFNet, enhances early Alzheimer's disease (AD) diagnosis by fusing structural and functional MRI data. It effectively addresses class imbalance and improves subtype classification accuracy.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Medical Diagnostics
Background:
- Early diagnosis of Alzheimer's disease (AD) is crucial but challenging due to subtle subtype differences and imbalanced medical imaging data.
- Current multimodal fusion methods for AD diagnosis using sMRI and rs-fMRI lack fine-grained cross-modal alignment and interaction.
Purpose of the Study:
- To propose a Deep Cross-Branch Multi-Modal Fusion Network (DCMFNet) for improved early AD diagnosis.
- To enhance cross-modal feature representation and mitigate class imbalance in AD diagnosis.
Main Methods:
- Preprocessing sMRI and rs-fMRI to extract ROI-based features, followed by dimension unification and normalization for cross-modal alignment.
- Utilizing a novel Deep Cross-branch Multi-modal Feature Fusion (DCMF) module with parallel and dual-pathway branches for comprehensive feature mining.
- Implementing a Transformer encoder for classification and introducing Logit Adjustment Cross-Entropy (LACE) loss to address class imbalance.
Main Results:
- DCMFNet demonstrated superior performance compared to traditional machine learning and state-of-the-art deep learning models across six binary AD subtype classification tasks.
- The DCMF module effectively improved cross-modal feature representation, while LACE loss successfully alleviated class imbalance.
- The proposed model achieved reliable early AD diagnosis on a private clinical dataset.
Conclusions:
- DCMFNet offers a robust multimodal fusion framework for early Alzheimer's disease diagnosis.
- The integration of DCMF module and LACE loss enhances diagnostic accuracy and addresses key challenges in medical imaging analysis for AD.
- This approach has the potential to reduce the diagnostic burden on healthcare professionals.