Discriminative multi-task feature selection for multi-modality classification of Alzheimer's disease

Area of Science:

• Neuroimaging
• Machine Learning
• Biomedical Data Analysis

Background:

• Multi-task feature selection is applied to classify Alzheimer's disease (AD) and mild cognitive impairment (MCI) using multi-modality data.
• Traditional methods often fail to fully exploit discriminative subject information, limiting classification performance.

Purpose of the Study:

• To propose a novel discriminative multi-task feature selection method for improved AD/MCI classification.
• To enhance the mining of discriminative information across multiple data modalities.

Main Methods:

• A linear regression model is trained for each modality, with group-sparsity regularization for joint feature selection.
• A discriminative regularization term using intra-class and inter-class Laplacian matrices is introduced.
• The method was evaluated on 202 subjects (AD, MCI, healthy controls) using MRI and FDG-PET data from ADNI.

Main Results:

• The proposed method significantly improved classification performance for AD/MCI.
• It demonstrated potential in discovering disease-related biomarkers for diagnosis.
• Performance was compared favorably against several state-of-the-art methods.

Conclusions:

• The discriminative multi-task feature selection method offers superior performance in multi-modality AD/MCI classification.
• The approach effectively leverages discriminative subject information and identifies potential biomarkers.
• This method advances the field of neuroimaging-based disease diagnosis.