Multimodal Alzheimer's disease classification through ensemble deep random vector functional link neural network.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) pathogenesis is complex, involving neuronal loss and characteristic brain pathologies.
• Early AD detection is crucial for intervention but challenged by data variability and single-modality research.
• Current methods often struggle with comprehensive staging due to incomplete or inconsistent data.

Purpose of the Study:

• To improve early detection and staging of Alzheimer's disease (AD) and mild cognitive impairment (MCI).
• To address limitations of single-modality data by integrating multimodal information (clinical, genetic).
• To evaluate the efficacy of deep learning models, specifically random vector functional link (RVFL) networks, for AD diagnosis.

Main Methods:

• Integration of multimodal data, including clinical and genetic information.
• Application of deep learning (DL) models, with a focus on ensemble deep RVFL (edRVFL) networks.
• Utilization of advanced data imputation techniques, such as Winsorized-mean (Wmean), to handle data incompleteness.

Main Results:

• The edRVFL model demonstrated superior performance in detecting early AD stages.
• Achieved high diagnostic metrics: 98.8% accuracy, 98.3% precision, 98.4% recall, and 98.2% F1-score.
• Outperformed traditional machine learning models (SVM, Random Forests, Decision Trees) in AD detection.

Conclusions:

• Integrating multimodal data with advanced DL techniques significantly enhances early AD detection.
• Ensemble deep RVFL models, coupled with effective imputation, offer a powerful approach for AD and MCI diagnosis.
• This study highlights the potential of sophisticated computational methods for improving neurological disorder diagnostics.