Compartmental sparse feature selection method for Alzheimer's disease identification.

Area of Science:

• Medical Imaging
• Neuroscience
• Machine Learning

Background:

• High-dimensional magnetic resonance imaging (MRI) data presents challenges for computer-aided Alzheimer's disease (AD) diagnosis.
• Existing feature selection methods aim to reduce dimensionality but can be improved for complex brain data.

Purpose of the Study:

• To present a novel compartmental sparse feature selection method for AD identification using T1-weighted MRI data.
• To improve the accuracy and efficiency of AD diagnosis through advanced feature selection techniques.

Main Methods:

• The proposed method partitions atlas-based regions-of-interest (ROIs) from MRI data into compartments.
• It employs sparse principal component analysis (SPCA) for local feature dimension estimation and selection within compartments.
• An elastic-net logistic regression (ENLR) classifier is used for compartmental classification.

Main Results:

• The compartmental approach demonstrated improved classification performance, particularly for smaller ROIs.
• The method achieved high computational efficiency in processing high-dimensional MRI data.
• Experimental results validate the effectiveness of the proposed feature selection strategy.

Conclusions:

• The compartmental sparse feature selection method offers a promising approach for accurate and efficient AD diagnosis.
• This technique effectively handles the complexity of high-dimensional MRI data for neurological disorder identification.
• The study highlights the potential of localized feature selection in improving diagnostic performance for Alzheimer's disease.