Contrast-free identification of glioma blood-brain barrier status via generative diffusion AI and non-contrast MRI - PubMed
4 hours ago
- #glioma
- #blood-brain barrier
- #generative AI
- Non-contrast MRI is used for preoperative glioma diagnosis and treatment planning, avoiding gadolinium-based contrast agents (GBCA) which can cause adverse effects.
- Generative AI models can synthesize contrast-enhanced images from non-contrast images, but conventional models face limitations due to tumor heterogeneity and synthesis errors.
- The study introduces a contrast-free BBB status identification model (CBSI) using non-contrast MRI and generative diffusion AI, achieving high accuracy (AUC = 81.31%).
- CBSI outperforms models using only non-contrast MRI (AUC = 72.76%) and performs comparably to T1Gd MRI models (AUC = 88.68%) in external tests.
- Validation on public datasets (BraTS-Africa and BraTS-GLI) confirms CBSI's generalizability for BBB status identification.
- CBSI improves glioma segmentation and grading accuracy by providing precise BBB status from synthetic T1Gd images, eliminating GBCA risks and enhancing clinical workflows.