Related Experiment Video
Updated: Oct 20, 2025
14:27
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
15.9K
Biomarker Extraction Based on Subspace Learning for the Prediction of Mild Cognitive Impairment Conversion
Ying Li1,2, Yixian Fang3, Jiankun Wang4
1Key Laboratory of TCM Data Cloud Service in Universities of Shandong, Shandong Management University, Jinan 250357, China.
Biomed Research International
|September 13, 2021
Summary
This study introduces novel biomarkers using subspace learning from brain MRI scans to better identify individuals with progressive mild cognitive impairment (MCI), aiding early Alzheimer's disease (AD) detection. The new method effectively distinguishes progressive MCI from stable MCI, outperforming existing approaches.
Area of Science:
- Neuroimaging
- Biomarker Discovery
- Machine Learning
Background:
- Accurate prediction of progressive mild cognitive impairment (MCI) is crucial for preventing Alzheimer's disease (AD).
- Extracting effective biomarkers from multivariate structural magnetic resonance imaging (MRI) data to differentiate progressive MCI from stable MCI remains challenging.
- Leveraging information from AD and normal control (NC) subjects can potentially improve MCI conversion prediction.
Purpose of the Study:
- To develop novel biomarkers for predicting MCI conversion to AD.
- To accurately differentiate progressive MCI from stable MCI using multivariate structural MRI data.
- To combine subspace learning with AD and NC subject information for enhanced biomarker extraction.
Main Methods:
- Developed novel biomarkers by combining subspace learning with AD and NC subject data.
- Learned projection matrices to map multivariate structural MRI data into a common label subspace.
- Applied self-weighted operation and weighted fusion on common subspace features for biomarker extraction.
- Validated biomarkers on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
Main Results:
- Proposed biomarkers significantly outperformed competing methods in discriminating between progressive MCI and stable MCI.
- The improvement in prediction accuracy was classifier-independent.
- Incorporating AD and NC subject information proved beneficial for predicting MCI conversion to AD.
- Demonstrated a promising representation of high-dimensional MRI data for MCI conversion prediction.
Conclusions:
- Novel subspace learning-based biomarkers show superior performance in predicting MCI conversion to AD.
- The developed method effectively extracts predictive features from high-dimensional MRI data.
- The findings highlight the utility of integrating information from related neurological states (AD, NC) for improved MCI prediction.
