Ensemble Deep Random Vector Functional Link Network Using Privileged Information for Alzheimer's Disease Diagnosis.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Medical Informatics

Background:

• Alzheimer's disease (AD) diagnosis requires advanced methods for early detection.
• Deep learning models show promise but face challenges like local minima and slow convergence.
• Existing deep random vector functional link network (RVFL) models cannot utilize privileged information.

Purpose of the Study:

• To develop deep RVFL models capable of incorporating privileged information for enhanced AD diagnosis.
• To propose a novel ensemble deep RVFL+ with learning using privileged information (LUPI) framework (edRVFL+) for improved classification accuracy and generalization.
• To introduce a new method for generating privileged information using distinct activation functions for normal and privileged data streams.

Main Methods:

• Incorporated Learning Using Privileged Information (LUPI) into deep RVFL, creating the dRVFL+ model.
• Developed an ensemble version, edRVFL+, optimizing a single network to generate an ensemble through random projections.
• Utilized different activation functions for normal and privileged information processing to generate distinct privileged information.

Main Results:

• Both dRVFL+ and edRVFL+ models demonstrated efficient utilization of privileged information, leading to better generalization performance.
• Experimental results confirmed the superiority of the proposed dRVFL+ and edRVFL+ models over baseline approaches in Alzheimer's disease diagnosis.
• The novel approach of generating separate privileged information using distinct activation functions proved effective.

Conclusions:

• The proposed dRVFL+ and edRVFL+ models offer a significant advancement in leveraging privileged information for deep learning-based AD diagnosis.
• The edRVFL+ model, in particular, shows potential for clinical application due to its enhanced classification accuracy and robustness.
• This work pioneers the integration of LUPI with deep RVFL architectures and introduces a novel method for privileged information generation.