Semi-supervised manifold learning with affinity regularization for Alzheimer's disease identification using positron...

Area of Science:

• Medical Imaging and Diagnostics
• Machine Learning in Healthcare
• Neuroscience Research

Background:

• Dementia, particularly Alzheimer's disease (AD), poses a significant global health challenge for aging populations.
• Current computer-aided dementia diagnosis methods often rely on supervised learning, requiring extensive labeled medical imaging datasets which are difficult to obtain in clinical settings.
• The potential of utilizing large unlabeled image datasets to improve dementia classification performance is recognized.

Purpose of the Study:

• To develop a novel semi-supervised dementia classification method that effectively utilizes unlabeled medical images.
• To improve the accuracy and practicality of computer-aided diagnosis for dementia and Alzheimer's disease.
• To evaluate the proposed method against state-of-the-art techniques using positron emission tomography (PET) data.

Main Methods:

• A semi-supervised dementia classification approach based on random manifold learning with affinity regularization was proposed.
• Spatial features were extracted from PET images to build an unsupervised random forest.
• The random forest was used to regularize the manifold learning objective function for classification of Alzheimer's disease (AD) and normal controls (NC).

Main Results:

• The experimental results demonstrated that incorporating unlabeled images significantly improves dementia classification performance.
• The proposed semi-supervised method outperformed the state-of-the-art Laplacian Support Vector Machine (LapSVM) on the tested dataset.
• The study validates the utility of semi-supervised learning in enhancing the accuracy of medical image-based dementia diagnosis.

Conclusions:

• Semi-supervised learning, particularly with the proposed random manifold learning approach, offers a promising avenue for improving dementia classification accuracy.
• The method's ability to leverage unlabeled data makes it more practical for real-world clinical applications compared to purely supervised methods.
• This research contributes to the advancement of AI-driven diagnostic tools for neurodegenerative diseases like Alzheimer's disease.