Elastic Net based sparse feature learning and classification for Alzheimer's disease identification.

Area of Science:

• Neuroimaging
• Biostatistics
• Machine Learning

Background:

• Magnetic Resonance Imaging (MRI) is crucial for studying neuronal changes.
• High-dimensional data in disease identification presents challenges like low sample size and high variable correlation.
• Sparse feature learning is vital for overcoming these challenges.

Purpose of the Study:

• To apply sparse Elastic Net (EN) for feature extraction and Alzheimer's disease (AD) identification using MRI data.
• To compare the effectiveness of EN against Principal Component Analysis (PCA).

Main Methods:

• Utilized sparse Elastic Net (EN) for dimension reduction and feature selection.
• Developed a unified formulation for dimension reduction and classification.
• Compared EN with PCA, highlighting EN's ability to handle correlated variables and group selection.

Main Results:

• EN demonstrated effectiveness in identifying Alzheimer's disease.
• EN outperformed PCA in scenarios with fewer samples and highly correlated variables.
• The method automatically selected relevant variable groups sharing biological phenomena.

Conclusions:

• Sparse Elastic Net is an effective method for Alzheimer's disease identification from MRI data.
• EN offers advantages over PCA by handling variable correlations and enabling automatic group selection.
• The integrated approach simplifies feature engineering and improves diagnostic accuracy.