Related Experiment Video
Updated: Jun 6, 2026
05:02
Modeling Brain Metastasis Via Tail-Vein Injection of Inflammatory Breast Cancer Cells
Published on: February 4, 2021
CT-to-FLAIR Style Transfer Improves Perilesional Edema Conspicuity in Brain Metastases
Dylan G Hsu1, Adhithya Narayanan1, Bala McRae-Posani2
1From the Department of Medical Physics (D.G.S.), Radiology (A.N., B.M.-P., O.Y., J.S., Akash S., V.H., Atin S., L.K., S.H., N.S., M.T.S., J.T., R.Y., A.H., H.S., J.N.S.), Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA.
AJNR. American Journal of Neuroradiology
|June 4, 2026
Summary
Generative adversarial networks (GANs) can create synthetic T2 FLAIR MRI scans from CT images, significantly improving the visibility of brain metastases. This AI-driven approach enhances lesion conspicuity compared to standard CT scans.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Brain metastases detection and characterization rely heavily on MRI, particularly T2 FLAIR sequences.
- Non-contrast CT is often the initial imaging modality, but it has limitations in visualizing certain lesions.
- Enhancing lesion conspicuity on CT could improve diagnostic accuracy and patient management.
Purpose of the Study:
- To assess the efficacy of generative adversarial networks (GANs) in synthesizing T2 FLAIR-like MR images from non-contrast CT.
- To determine if synthesized T2 FLAIR images improve the conspicuity of brain metastases compared to CT alone.
Main Methods:
- Retrospective analysis of 321 patients with paired non-contrast CT and T2 FLAIR MR images.
- Training GANs (PatchGAN discriminator, UNet++ generator) for CT-to-synthetic-FLAIR style transfer.
- Quantitative evaluation using MAE, MSE, SSIM, and qualitative assessment by 12 neuroradiologists on image quality and lesion conspicuity.
Main Results:
- UNet++ achieved superior image quality metrics (SSIM=0.9119, MAE=0.1232, MSE=0.0201).
- Synthetic MR images showed significantly improved lesion conspicuity over CT (p<0.001) but slightly less than real MR (p<0.001).
- Reader ratings indicated high image quality for synthetic MR (4.10/5★), comparable to real MR (4.28/5★), and superior conspicuity over CT (3.16/5★).
Conclusions:
- GAN-based style transfer can effectively synthesize T2 FLAIR MR images from non-contrast CT.
- Synthesized T2 FLAIR images enhance the conspicuity of brain metastases edema, offering a valuable adjunct to CT imaging.
- This AI technique shows promise for improving the detection and assessment of brain metastases.
