Biologically inspired hybrid model for Alzheimer's disease classification using structural MRI in the ADNI dataset

Area of Science:

• Artificial Intelligence
• Neuroscience
• Medical Imaging

Background:

• Alzheimer's disease (AD) diagnosis is challenging due to subtle early-stage neuroanatomical changes.
• Accurate staging of AD is crucial for timely intervention and treatment.

Purpose of the Study:

• To develop and evaluate a hybrid CNN-SNN model for classifying Alzheimer's disease, mild cognitive impairment, and cognitively normal subjects using structural MRI data.
• To leverage the strengths of CNNs for spatial feature extraction and SNNs for temporal dynamics processing in neurodegeneration.

Main Methods:

• A hybrid CNN-SNN architecture was designed, integrating CNNs for spatial feature learning and SNNs with leaky integrate-and-fire neurons for temporal processing.
• The model was trained on structural MRI data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) for a three-class classification task.
• Optimization involved MSE loss, L2 regularization, Adam optimizer, and early stopping to ensure generalization.

Main Results:

• The hybrid CNN-SNN model achieved high accuracy in classifying AD stages.
• An ablation study revealed the SNN component's critical role, with its removal significantly decreasing accuracy from 99.58% to 75.67%.
• The model demonstrated the ability to capture complex spatiotemporal patterns indicative of neurodegeneration.

Conclusions:

• Hybrid CNN-SNN architectures offer a computationally efficient and biologically plausible approach for AD diagnosis using neuroimaging data.
• This approach has the potential to enhance early detection and stratification of neurodegenerative diseases.
• The findings support the integration of neuromorphic principles into healthcare systems for improved diagnostics.