Relational-Regularized Discriminative Sparse Learning for Alzheimer's Disease Diagnosis

Area of Science:

• Neuroscience
• Machine Learning
• Medical Imaging Analysis

Background:

• Accurate early diagnosis and prognosis of Alzheimer's disease (AD) are crucial.
• Identifying informative features is key for effective AD detection.
• Multimodal data integration offers potential for improved diagnostic accuracy.

Purpose of the Study:

• To develop a novel discriminative sparse learning method with relational regularization for AD.
• To jointly predict clinical scores and classify AD disease stages using multimodal features.
• To enhance the accuracy of Alzheimer's disease diagnosis and prognosis.

Main Methods:

• Proposed a discriminative sparse learning method incorporating relational regularization.
• Utilized a discriminative learning technique to enhance class-specific differences and geometric information for feature selection.
• Incorporated two types of relational information for similarity learning among features and subjects.
• Mapped features into a target space for informative feature identification via sparse learning.
• Designed a unique loss function combining discriminative learning and relational regularization.

Main Results:

• Achieved high classification accuracy: 94.68% for AD vs. Normal Controls (NC), 80.32% for Mild Cognitive Impairment (MCI) vs. NC, and 74.58% for progressive MCI vs. stable MCI.
• Demonstrated remarkable performance in predicting clinical scores and identifying classification labels.
• Experimental results based on 805 subjects from the AD neuroimaging initiative database.
• The proposed method showed superiority over existing state-of-the-art methods.

Conclusions:

• The developed method is effective for Alzheimer's disease diagnosis and prognosis.
• The combination of discriminative learning and relational regularization significantly improves performance.
• The approach holds promise for early and accurate detection of Alzheimer's disease and its progression.