Bio-inspired deep learning-personalized ensemble Alzheimer's diagnosis model for mental well-being

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Current Alzheimer's Diagnosis (AD) models often overlook individual patient differences, impacting diagnostic accuracy.
• A need exists for personalized classification strategies that account for sample-specific variations in disease progression.

Purpose of the Study:

• To introduce a novel personalized dynamic ensemble convolution neural network (PDECNN) for improved Alzheimer's Diagnosis (AD).
• To develop a model capable of dynamically adjusting to sample-specific brain degeneration patterns.

Main Methods:

• Proposed a personalized dynamic ensemble convolution neural network (PDECNN) model.
• Employed an attention mechanism to evaluate degeneration in specific brain regions.
• Dynamically selected and integrated brain region features based on degeneration levels.

Main Results:

• The PDECNN model demonstrated improved classification accuracy by 4%, 11%, and 8% across different metrics.
• Identified degraded brain regions showed high consistency with known clinical manifestations of Alzheimer's disease.
• The model successfully adapted to variations in brain area degeneration across different samples.

Conclusions:

• The PDECNN model offers a personalized approach to Alzheimer's Diagnosis (AD) by accounting for individual sample distinctiveness.
• This method enhances diagnostic importance by identifying sample-specific degraded brain areas.
• The PDECNN model represents a significant advancement in AI-driven diagnostic tools for neurodegenerative diseases.