Ensemble sparse classification of Alzheimer's disease.

Area of Science:

• Neuroimaging analysis
• Machine learning for medical diagnosis
• Pattern classification in high-dimensional data

Background:

• Support vector machines (SVM) and other pattern classification methods are used for Alzheimer's disease (AD) and mild cognitive impairment (MCI) diagnosis via neuroimaging.
• Existing methods often rely on a single global classifier, which struggles with noise and small sample sizes common in neuroimaging data, limiting robustness and performance.
• Accurate and early diagnosis of AD and MCI is crucial for effective patient management and treatment strategies.

Purpose of the Study:

• To develop a more robust and accurate classification method for diagnosing Alzheimer's disease (AD) and mild cognitive impairment (MCI) using magnetic resonance imaging (MRI).
• To overcome the limitations of single global classifiers in handling noisy and small-sample neuroimaging datasets.
• To leverage local image features through an ensemble approach for improved diagnostic performance.

Main Methods:

• Proposed a local patch-based subspace ensemble method for neuroimaging data classification.
• Partitioned brain MRI images into local patches and constructed weak classifiers using subsets of these patches.
• Employed Sparse Representation-based Classification (SRC) to build individual weak classifiers, combining multiple classifiers for the final decision.

Main Results:

• Achieved 90.8% accuracy and 94.86% AUC for Alzheimer's disease (AD) classification on a dataset of 652 subjects from the ADNI database.
• Attained 87.85% accuracy and 92.90% AUC for mild cognitive impairment (MCI) classification.
• Demonstrated superior performance compared to state-of-the-art methods for AD/MCI classification using MRI data.

Conclusions:

• The proposed local patch-based subspace ensemble method offers a robust and accurate approach for diagnosing AD and MCI from MRI.
• Ensemble learning with local image features effectively addresses challenges posed by noise and limited sample sizes in neuroimaging.
• The method shows significant promise for clinical application in the early detection of Alzheimer's disease and its prodromal stages.