Related Experiment Video
Updated: Mar 15, 2026
12:18
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
8.2K
Multimodal Classification of Mild Cognitive Impairment Based on Partial Least Squares
Pingyue Wang1, Kewei Chen2, Li Yao1,3
1National Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University, Beijing, China.
Journal of Alzheimer'S Disease : JAD
|August 29, 2016
Summary
This study shows that combining MRI, FDG-PET, and florbetapir-PET scans using a multimodal partial least squares (PLS) model can accurately predict Alzheimer's disease conversion in mild cognitive impairment patients. This approach aids early diagnosis.
Area of Science:
- Neuroscience
- Medical Imaging
- Biostatistics
Background:
- Early identification of mild cognitive impairment (MCI) conversion to Alzheimer's disease (AD) is crucial for timely intervention.
- Neuroimaging plays a vital role in classifying and predicting MCI progression.
- Existing methods often rely on single imaging modalities or less advanced analytical techniques.
Purpose of the Study:
- To develop and evaluate a multimodal neuroimaging approach for discriminating MCI converters (MCI-c) from MCI non-converters (MCI-nc).
- To compare the efficacy of partial least squares (PLS) models using combined MRI, FDG-PET, and florbetapir-PET data against single-modality approaches.
- To assess the added value of incorporating clinical scores into the multimodal PLS model for enhanced diagnostic accuracy.
Main Methods:
- Utilized data from 64 MCI-c and 65 MCI-nc participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.
- Employed two types of PLS models: informed PLS and agnostic PLS, integrating magnetic resonance imaging (MRI), 18F-fluorodeoxyglucose PET (FDG-PET), and 18F-florbetapir PET (florbetapir-PET) data.
- Compared the diagnostic performance of the multimodal PLS models with single-modality models and other machine learning algorithms like Support Vector Machine (SVM) and Random Forest (RF).
Main Results:
- The three-modality informed PLS model achieved 81.40% accuracy, 79.69% sensitivity, and 83.08% specificity.
- The three-modality agnostic PLS model demonstrated superior classification performance compared to two-modality models.
- Integrating clinical test scores (ADAS-cog) with the agnostic PLS model (florbetapir-PET as independent, FDG-PET and MRI as dependent) yielded optimal results: 86.05% accuracy, 81.25% sensitivity, and 90.77% specificity.
- PLS models exhibited greater diagnostic power than SVM and RF.
Conclusions:
- Multimodal neuroimaging analysis using PLS is effective in differentiating MCI converters from non-converters.
- Combining MRI, FDG-PET, and florbetapir-PET offers enhanced predictive capabilities for Alzheimer's disease progression.
- The developed multimodal PLS model shows significant potential for early and accurate diagnosis of Alzheimer's disease in individuals with MCI.
