Related Experiment Video
Updated: May 10, 2026
12:50
Lesion Explorer: A Video-guided, Standardized Protocol for Accurate and Reliable MRI-derived Volumetrics in Alzheimer's Disease and Normal Elderly
Published on: April 14, 2014
Locally linear embedding (LLE) for MRI based Alzheimer's disease classification
Xin Liu1, Duygu Tosun, Michael W Weiner
1Center of Imaging of Neurodegenerative Disease, VA Medical Center and the Department of Radiology and Biomedical Imaging University of California, San Francisco, CA, USA.
Neuroimage
|June 25, 2013
Summary
Local linear embedding (LLE) improved Alzheimer's disease (AD) prediction from brain MRI scans. This machine learning method enhances the analysis of complex MRI data for earlier and more accurate AD diagnosis.
Area of Science:
- Neuroimaging and Machine Learning
- Computational Neuroscience
- Biomedical Data Analysis
Background:
- Machine learning is increasingly applied to neuroimaging for predicting neurodegenerative diseases like Alzheimer's disease (AD).
- Extracting meaningful patterns from complex, multivariate brain MRI data for prediction remains a challenge.
- Effective feature representation is crucial for accurate diagnostic and prognostic models in AD.
Purpose of the Study:
- To evaluate the effectiveness of Local Linear Embedding (LLE) in improving predictions of Alzheimer's disease (AD) using structural MRI data.
- To assess LLE's ability to transform high-dimensional MRI features into a more informative, lower-dimensional space.
- To compare classification performance using LLE-embedded features versus original MRI features for predicting MCI conversion to AD.
Main Methods:
- Applied unsupervised Local Linear Embedding (LLE) to transform regional brain volume and cortical thickness MRI data.
- Utilized the embedded features to train various classifiers (logistic regression, SVM, LDA) for AD prediction.
- Tested the framework on 413 individuals from the Alzheimer's Disease Neuroimaging Initiative (ADNI) with 3-year follow-up.
Main Results:
- Classifications using LLE-embedded MRI features significantly outperformed those using original features (p<0.05).
- LLE improved classification performance across multiple algorithms, demonstrating broad applicability.
- LLE notably enhanced the prediction of Mild Cognitive Impairment (MCI) converters to AD (accuracy 0.68) compared to chance (accuracy 0.56).
Conclusions:
- Local Linear Embedding (LLE) is a highly effective tool for enhancing classification accuracy in Alzheimer's disease (AD) studies using multivariate MRI data.
- The improved prediction of MCI conversion to AD has significant implications for clinical management and clinical trial targeting.
- LLE facilitates better utilization of complex neuroimaging data for disease prediction and early intervention strategies.
