Related Experiment Video
Updated: Aug 4, 2025
12:18
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
7.6K
Multi-level fusion network for mild cognitive impairment identification using multi-modal neuroimages
Haozhe Xu1,2,3, Shengzhou Zhong1,2,3, Yu Zhang1,2,3
1School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, People's Republic of China.
Physics in Medicine and Biology
|April 5, 2023
Summary
This study introduces a new deep learning model for identifying mild cognitive impairment (MCI) using multi-modal neuroimaging. The model effectively captures local and global dependencies for improved MCI diagnosis and prediction.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Medical Diagnostics
Background:
- Mild cognitive impairment (MCI) precedes Alzheimer's disease (AD), necessitating early detection.
- Current deep learning methods for MCI identification often overlook dependencies within multi-modal neuroimage features.
- Existing approaches may inadequately integrate modality-sharable and modality-specific information.
Purpose of the Study:
- To develop an accurate multi-level fusion network for mild cognitive impairment identification using multi-modal neuroimages.
- To address limitations in previous methods by modeling local feature dependencies and incorporating both sharable and specific information.
Main Methods:
- Proposed a multi-level fusion network with local and dependency-aware global representation learning stages.
- Extracted multi-pair patches from multi-modal neuroimages at corresponding positions.
- Employed dual-channel sub-networks with modality-specific branches and sine-cosine fusion modules for local feature learning.
- Integrated long-range dependencies among local representations for global feature learning.
Main Results:
- Achieved superior performance on ADNI-1/ADNI-2 datasets for MCI identification.
- Demonstrated high accuracy (0.802), sensitivity (0.821), and specificity (0.767) in MCI diagnosis.
- Showcased strong results in predicting MCI conversion (accuracy: 0.849, sensitivity: 0.841, specificity: 0.856).
Conclusions:
- The proposed multi-level fusion network effectively identifies MCI using multi-modal neuroimaging.
- The method shows significant potential for predicting MCI conversion and identifying relevant brain regions.
- Experimental results validate the feasibility and superiority of the developed approach.
