Synergistic medical genetic evolutionary optimization and deep convolutional generative augmentation with SHAP-driven...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Neuroscience

Background:

• Accurate early-stage Alzheimer's disease (AD) diagnosis is vital for effective treatment but hindered by data imbalance, model interpretability, and computational costs.
• Existing diagnostic models often struggle with imbalanced datasets and lack transparency, impacting clinical adoption.

Purpose of the Study:

• To develop a novel, end-to-end diagnostic framework for early and accurate Alzheimer's disease detection.
• To address challenges of data imbalance, low model interpretability, and high computational cost in AD diagnosis.

Main Methods:

• Utilized a Medical Genetic Algorithm (MedGA)-optimized Convolutional Neural Network (CNN) integrated with a Deep Convolutional Generative Adversarial Network (DCGAN) for synthetic MRI generation.
• Employed SHapley Additive exPlanations (SHAP) for model interpretability, analyzing key brain regions like the hippocampus and amygdala.
• Trained and validated the framework on the Open Access Series of Imaging Studies (OASIS) dataset, focusing on four AD stages.

Main Results:

• DCGAN generated 700 synthetic MRIs, improving recall by 10% and balancing the dataset for the Moderate Dementia class (SSIM=0.92).
• MedGA optimized CNN hyperparameters, reducing network complexity by 20% while maintaining 97% testing accuracy.
• SHAP analysis revealed the hippocampus and amygdala as critical for classification, enhancing interpretability by 25% and clinician confidence.

Conclusions:

• The proposed framework demonstrates superior predictive performance and explainability compared to state-of-the-art methods for Alzheimer's disease diagnosis.
• This integrated approach offers a powerful tool for early AD categorization, promising improved diagnostic precision in healthcare settings.