Ensemble support vector machine classification of dementia using structural MRI and mini-mental state examination.

Area of Science:

• Neuroimaging
• Machine Learning
• Computational Neuroscience

Background:

• The International Challenge for Automated Prediction of MCI from MRI data aimed to compare machine learning algorithms for classifying normal controls (NC), mild cognitive impairment (MCI), converting MCI (cMCI), and Alzheimer's disease (AD).
• Accurate classification of these neurological conditions is crucial for timely intervention and treatment.

Purpose of the Study:

• To evaluate the effectiveness of an ensemble of support vector machines (SVMs) for multi-class classification of dementia subtypes using MRI data.
• To assess the benefits of combining bagging and feature selection techniques with SVMs for improved diagnostic accuracy.

Main Methods:

• An ensemble of SVM classifiers was developed, incorporating bagging without replacement and feature selection.
• Both linear and radial basis function (RBF) kernels were utilized within the SVM framework.
• Post-challenge analysis explored the impact of minimum feature selection and increased ensemble size on performance.

Main Results:

• The ensemble SVM achieved multi-class classification accuracies of 55.6% (linear kernel) and 55.0% (RBF kernel) on the challenge test set, securing third place.
• The most frequently selected MRI features were the volumes of the left presubiculum and right subiculum.
• Further analysis demonstrated that optimizing feature selection and ensemble size improved accuracy to 59.1%.

Conclusions:

• Ensemble methods, particularly those employing bagging and feature selection, enhance the performance of SVM classifiers in dementia classification.
• This approach yielded competitive classification accuracies in the International Challenge for Automated Prediction of MCI from MRI data.
• The findings highlight the potential of advanced machine learning techniques for automated neurological disorder prediction.