Related Experiment Video
Updated: Jun 27, 2025
14:15
Multicolor 3D Printing of Complex Intracranial Tumors in Neurosurgery
Published on: January 11, 2020
7.1K
Synthesizing 3D Multi-Contrast Brain Tumor MRIs Using Tumor Mask Conditioning
Nghi C D Truong1, Chandan Ganesh Bangalore Yogananda1, Benjamin C Wagner1
1Department of Radiology, UT Southwestern Medical Center, Texas, USA.
Summary
This study introduces a novel method using generative AI to create synthetic 3D multi-contrast brain tumor MRI data, addressing data scarcity for deep learning models.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Deep Learning
Background:
- Data scarcity and imbalance challenge deep learning model training on medical images like brain tumor MRIs.
- Generative AI offers a promising solution for synthetic MRI data generation, enhancing training datasets.
Purpose of the Study:
- To adapt latent diffusion models for generating 3D multi-contrast brain tumor MRI data.
- To address limitations in existing methods by generating 3D multi-contrast samples guided by tumor masks.
Main Methods:
- Developed a framework with a 3D autoencoder for compression and a conditional 3D Diffusion Probabilistic Model (DPM).
- Integrated a conditional module in the DPM's UNet to utilize tumor masks for guided generation.
- Trained models on The Cancer Genome Atlas (TCGA) and UTSW datasets.
Main Results:
- Successfully generated high-quality 3D multi-contrast brain tumor MRI samples.
- Generated images demonstrated accurate tumor localization guided by the input condition mask.
- Evaluated image quality using the Fréchet Inception Distance (FID) score.
Conclusions:
- The proposed method effectively generates diverse and high-quality 3D multi-contrast brain tumor MRI data.
- This approach has the potential to significantly mitigate brain tumor data scarcity.
- Improved data availability can enhance the performance of deep learning models in neuro-oncology.
