DeepSymNet

DeepSymNet quantifies longitudinal neurodegenerative progression from raw 3D structural MRI voxel data using an end-to-end deep learning architecture to identify anatomical changes relevant to Alzheimer's disease.

Key Features:

• Data-Driven Architecture: Analyzes raw voxel data from 3D structural MRI scans to identify longitudinal changes between two time points without predefined regions of interest.
• End-to-End Training: Trains the deep learning model end-to-end on imaging data, removing dependence on preprocessing steps such as non-linear registration and iterative surface deformations.
• Atlas- and Registration-Free Analysis: Eliminates reliance on atlases, predefined ROIs, and complex non-linear registration pipelines.
• Efficiency and Speed: Significantly reduces processing time compared to Freesurfer longitudinal pipelines while maintaining comparable accuracy in identifying AD progression.
• Statistical Performance: Demonstrated statistically significant improvement over other voxel-based methods when evaluated using the Alzheimer's Disease Neuroimaging Initiative (ADNI) database and distinguishes healthy subjects from those with mild cognitive impairment (MCI).
• Anatomical Insights via Epsilon Layer-wise Propagation: Uses epsilon layer-wise propagation analysis to highlight regions including the pallidum, putamen, and superior temporal gyrus associated with AD progression.

Scientific Applications:

• Early diagnosis and monitoring of Alzheimer's disease: Supports detection and longitudinal monitoring of Alzheimer's disease progression and differentiation of healthy controls from MCI.
• Longitudinal neuroimaging studies: Enables longitudinal analyses by minimizing preprocessing-induced error propagation and bypassing atlas-based registration.

Methodology:

End-to-end deep learning on raw 3D structural MRI voxels to measure longitudinal changes between two time points; epsilon layer-wise propagation analysis for region attribution; evaluation against voxel-based methods using the ADNI database; processing-time comparison with Freesurfer longitudinal pipelines.