[Supervised locally linear embedding for magnetic resonance imaging based Alzheimer's disease classification]

Area of Science:

• Neuroscience
• Machine Learning
• Medical Imaging

Context:

• Early classification of Alzheimer's disease (AD) is challenging due to high-dimensional neuroimaging data.
• Conventional linear feature extraction methods struggle to identify discriminative information for accurate classification of unlabeled samples.

Purpose:

• To reduce redundant features and enhance recognition accuracy for early AD detection.
• To apply the supervised locally linear embedding (SLLE) algorithm for transforming regional brain volume and cortical thickness data into a lower-dimensional space.

Summary:

• The study utilized the SLLE algorithm with a distance correction term to compute nearest neighbors and reconstruct data in a lower-dimensional space.
• Feature extraction algorithms including SLLE, PCA, NMMP, and LLE were combined with SVM for classifying cognitive states (CN, sMCI, aMCI, AD).
• SLLE combined with SVM demonstrated improved accuracy, sensitivity, and specificity in classifying mild cognitive impairment subtypes compared to LLE-SVM and SVM alone.

Impact:

• The SLLE-SVM combination shows enhanced effectiveness for the early diagnosis of Alzheimer's disease.
• Achieved a 1.08% accuracy improvement over LLE-SVM and a 7.91% improvement over SVM in classifying sMCI and aMCI.
• Highlights the potential of SLLE for improving diagnostic accuracy in neurodegenerative disease research.