Probability distribution function-based classification of structural MRI for the detection of Alzheimer's disease

Area of Science:

• Neuroimaging
• Machine Learning
• Medical Diagnostics

Background:

• High-dimensional classification is crucial for Alzheimer's disease (AD) diagnosis using neuroimaging data.
• Effective feature selection from complex datasets like magnetic resonance imaging (MRI) remains a significant challenge.

Purpose of the Study:

• To introduce a novel statistical feature reduction and selection method for high-dimensional structural MRI (sMRI) data.
• To develop an automatic computer-aided diagnosis (CAD) technique for early AD detection.
• To compare the proposed method with existing feature selection techniques.

Main Methods:

• Utilized voxel-based morphometry (VBM) to analyze gray matter differences between AD patients and healthy controls (HCs) in 3-Tesla 3D T1-weighted MRI data.
• Extracted features from voxel clusters identified by VBM, defining volumes of interest (VOIs).
• Employed a novel probability distribution function (PDF) based statistical feature selection process on VOIs and assessed performance using support vector machine (SVM) classifiers and 10-fold cross-validation on 130 AD and 130 HC subjects from the ADNI dataset.

Main Results:

• The proposed PDF-based feature selection method effectively identified statistical patterns in high-dimensional sMRI data.
• The method demonstrated high competitiveness compared to standard feature selection techniques like partial least squares (PLS).
• Achieved reliable classification of AD from sMRI data, indicating its potential for early detection.

Conclusions:

• The PDF-based feature selection approach is a reliable and effective technique for classifying Alzheimer's disease from high-dimensional sMRI data.
• This novel method shows significant promise for developing advanced computer-aided diagnosis systems for neurological disorders.
• The findings suggest the PDF approach is a competitive alternative to current state-of-the-art feature selection methods in neuroimaging analysis.