Gender-based Alzheimer's detection using ResNet-50 and binary dragonfly algorithm on neuroimaging.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder with a silent onset, making early diagnosis challenging.
• Existing research shows AD risk varies by gender, age, race, and ethnicity, yet gender-based studies are lacking.
• Accurate diagnosis of Alzheimer's requires sophisticated algorithms to detect subtle brain changes.

Purpose of the Study:

• To propose and evaluate a gender-based Alzheimer's detection model (GRDN) using functional magnetic resonance imaging (fMRI).
• To explore the impact of gender on Alzheimer's detection accuracy using machine learning.
• To enhance classifier performance on underrepresented groups through data balancing techniques.

Main Methods:

• Utilized fMRI data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
• Employed a generative adversarial network (GAN) for data balancing and a ResNet-50 architecture for feature extraction.
• Applied the binary dragonfly algorithm (BDA) for feature selection, followed by five machine learning classifiers.

Main Results:

• Increased feature set size correlated with improved classification accuracy.
• The fineKNN classifier achieved a high accuracy of 94.8% for the male group with a feature set of 450.
• The proposed model demonstrated consistent performance across different study groups, outperforming other models.

Conclusions:

• The GRDN model shows significant potential for gender-based Alzheimer's detection.
• Feature engineering and selection are crucial for enhancing diagnostic accuracy in Alzheimer's.
• Further research into gender-specific risk factors and diagnostic approaches is warranted.