Behavior Score-Embedded Brain Encoder Network for Improved Classification of Alzheimer Disease Using Resting State...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Neurology

Background:

• Accurate early detection of dementia, including Alzheimer Disease (AD) and Mild Cognitive Impairment (MCI), is crucial for effective patient treatment.
• Current clinical diagnosis relies on integrating psychological assessments with neuroimaging techniques like PET and MRI.
• Resting-state functional magnetic resonance imaging (fMRI) offers valuable insights into brain function relevant to cognitive decline.

Purpose of the Study:

• To develop and validate a novel deep learning framework for automatic classification of dementia.
• To integrate behavioral scores from psychological tests with fMRI data for enhanced diagnostic accuracy.
• To identify brain regions most discriminative between healthy controls and AD patients.

Main Methods:

• A Behavior Score-Embedded Encoder Network (BSEN) was developed, utilizing a 3D convolutional autoencoder architecture.
• BSEN was optimized using contrastive loss and integrated behavioral scores from the Mini-Mental State Examination (MMSE) and Clinical Dementia Rating (CDR).
• The framework was evaluated on two distinct datasets for classifying three groups: AD, MCI, and Healthy Control (HC).

Main Results:

• The BSEN-based classification framework achieved an overall recognition accuracy of 59.44% for the 3-class classification task.
• The study successfully extracted brain regions exhibiting significant differences between healthy controls and AD patients.
• The integration of behavioral scores improved the representation of fMRI data for classification.

Conclusions:

• The proposed BSEN framework demonstrates a promising approach for the early detection and classification of dementia using integrated neuroimaging and behavioral data.
• This method offers a potential tool to aid clinicians in diagnosing AD and MCI more accurately.
• Further research can explore refining the BSEN architecture and incorporating additional clinical data for improved diagnostic performance.