AI-Driven Synthetic Brain MRI for Regional Atrophy in Neurodegeneration
Overview
This study evaluates the performance of deep learning-generated synthetic 3D T1-weighted MRI sequences in detecting regional brain atrophy in Alzheimer’s disease (AD) and frontotemporal dementia (FTD). Synthetic images derived from various clinical input sequences showed strong volumetric correlations with acquired 3D T1 sequences and enabled automated classification of neurodegenerative patterns.
Background
Neurodegenerative diseases like AD and FTD are major causes of morbidity worldwide, with increasing prevalence expected. Brain MRI, particularly 3D T1-weighted sequences, is essential for assessing regional brain atrophy patterns characteristic of these disorders. However, 3D T1 sequences are often not acquired in routine clinical scans, limiting volumetric analyses. Deep learning methods such as SynthSR can generate synthetic high-resolution 3D T1 images from other MRI sequences, potentially overcoming this limitation.
Data Highlights
Sequence Type
Input Sequence
Resolution
Real T1-weighted (rT1)
Acquired
1 mm isotropic 3D
Synthetic T1-weighted (sT1)
3D FLAIR
1 mm isotropic 3D
Synthetic T1-weighted (sT1)
Axial FLAIR
4 mm
Synthetic T1-weighted (sT1)
Coronal T2
4 mm
Key Findings
SynthSR-generated synthetic 3D T1 images from various input sequences demonstrated high volumetric correlation with acquired 3D T1 sequences.
Regional volumetric measurements in disease-specific atrophy areas (hippocampus in AD, frontal and temporal lobes in FTD) were reliably reproduced in synthetic images.
Synthetic sequences enabled automated classification algorithms to distinguish AD and FTD patients from healthy controls effectively.
The proposed nomenclature system (e.g., sT13DFLAIR) facilitates clear differentiation between real and synthetic images and their input sources.
Defacing for anonymization had negligible impact on atrophy estimation, supporting data privacy compliance.
Clinical Implications
Deep learning-based synthetic 3D T1 MRI sequences can supplement missing volumetric reference scans in routine clinical practice, enabling reliable assessment of regional brain atrophy in neurodegenerative diseases. This approach may improve diagnostic workflows by allowing automated volumetric analyses and classification even when standard 3D T1 acquisitions are unavailable.
Conclusion
SynthSR-generated synthetic 3D T1-weighted images from diverse clinical MRI inputs provide a viable alternative to acquired 3D T1 sequences for volumetric assessment of regional brain atrophy in AD and FTD. This technology holds promise to enhance neurodegenerative disease diagnostics in routine clinical settings.
References
SynthSR Tool and FreeSurfer v7.3 -- Synthetic MRI Generation
Power Analysis Using G*Power 3.1 -- Sample Size Justification
PyDeface Anonymization Tool -- Impact on Atrophy Estimation
