Related Experiment Video
Updated: May 12, 2026
09:33
Diffusion Tensor Magnetic Resonance Imaging in the Analysis of Neurodegenerative Diseases
Published on: July 28, 2013
28.4K
Improved Dementia Prediction in Cerebral Small Vessel Disease Using Deep Learning-Derived Diffusion Scalar Maps From
Yutong Chen1, Daniel Tozer1, Rui Li1
1Department of Clinical Neuroscience, Stroke Research Group, University of Cambridge, United Kingdom (Y.C., D.T., R.L., H.S.M.).
Stroke
|August 15, 2024
Summary
Synthesize fractional anisotropy (FA) and mean diffusivity (MD) maps from T1 images using deep learning. This method enhances dementia prediction in small vessel disease when diffusion tensor imaging is unavailable.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Neurology
Background:
- Cerebral small vessel disease is a primary cause of vascular dementia.
- Diffusion tensor imaging (DTI) detects white matter damage better than conventional MRI but is time-consuming and not routine.
- Synthesizing DTI-derived scalar maps (FA/MD) from T1 images offers a potential solution.
Purpose of the Study:
- To develop a deep learning model for synthesizing FA/MD maps from T1 images.
- To evaluate the accuracy and generalizability of the synthesized maps.
- To assess the utility of synthesized maps in predicting dementia risk.
Main Methods:
- A deep learning model was trained on UK Biobank data (n=4998) with high white matter hyperintensity volumes.
- External validation was performed on four small vessel disease datasets (SCANS, RUN DMC, PRESERVE, NETWORKS) and UK Biobank normal controls.
- The model synthesized FA/MD maps from T1 images.
Main Results:
- Synthetic MD and FA maps closely resembled ground-truth maps (SSIM >0.89 for MD, >0.80 for FA).
- Dementia prediction accuracy using synthetic MD was comparable to ground truth (e.g., SCANS c-index: synthetic 0.821 vs. ground truth 0.822).
- Synthetic map prediction outperformed white matter hyperintensity volume in dementia prediction.
Conclusions:
- A fast and generalizable deep learning method was developed to synthesize FA/MD maps from T1 images.
- This approach improves dementia prediction accuracy in small vessel disease cases lacking DTI data.
- The method offers a valuable tool for clinical practice and research.
