Optimized Hybrid Deep Learning Framework for Early Detection of Alzheimer's Disease Using Adaptive Weight Selection

Area of Science:

• Artificial Intelligence
• Medical Imaging Analysis
• Neurological Disorders

Background:

• Alzheimer's disease (AD) is a progressive neurological disorder causing cognitive decline.
• Current diagnostic methods for AD often lack accuracy and efficiency.
• Early and precise diagnosis is critical for effective Alzheimer's disease intervention.

Purpose of the Study:

• To develop an enhanced hybrid deep learning framework for improved Alzheimer's disease diagnosis.
• To leverage the strengths of EfficientNetV2B3 and Inception-ResNetV2 models for superior classification accuracy.
• To address the limitations of conventional diagnostic techniques in identifying early-stage Alzheimer's disease.

Main Methods:

• A hybrid deep learning framework combining EfficientNetV2B3 and Inception-ResNetV2 was developed.
• An adaptive weight selection process using the Cuckoo Search optimization algorithm was employed.
• Neuroimaging data underwent pre-processing, followed by feature extraction and dynamic model weighting for optimized performance.

Main Results:

• The framework achieved a Scott's Pi agreement score of 0.9907, demonstrating exceptional diagnostic accuracy.
• The model showed superior performance in identifying early-stage Alzheimer's disease compared to existing methods.
• The results were validated on extensive neuroimaging datasets, confirming the framework's efficacy.

Conclusions:

• The proposed framework shows significant potential as a reliable tool for Alzheimer's disease identification.
• This hybrid approach mitigates shortcomings of conventional methods and current deep learning algorithms.
• The adaptive nature of the Cuckoo Search optimization enhances applicability across diverse neuroimaging datasets and diagnostic scenarios.