Related Experiment Video
Updated: Aug 14, 2026
Co-analysis of Brain Structure and Function using fMRI and Diffusion-weighted Imaging
Published on: November 8, 2012
POWDR: Pathology-Preserving Outpainting with Wavelet Diffusion for 3D MRI
Fei Tan1, Ashok Vardhan Addala1, Bruno Astuto Arouche Nunes1
1GE HealthCare, San Ramon, CA 94583, USA.
Abstract:
Background: Medical imaging datasets often suffer from class imbalance and limited availability of pathology-rich cases, which limit the performance of machine learning models for segmentation, classification, and vision-language tasks. We propose POWDR, a pathology-preserving outpainting framework for 3D MRI that uses real pathological regions as conditioning evidence while generating anatomically plausible surrounding tissue. Methods: Our approach leverages wavelet-domain conditioning to enhance high-frequency detail and mitigate blurring common in latent diffusion models. We introduce a random connected mask training strategy to reduce conditioning-induced collapse and improve diversity outside the lesion. POWDR is evaluated on brain MRI using BraTS datasets and extended to knee MRI to assess applicability beyond brain imaging. Results: Quantitative metrics (FID, MS-SSIM, LPIPS) were used to assess image realism. Random connected mask training improved diversity, reducing cosine similarity from 0.9947 to 0.9580 and increasing KL divergence from 0.00026 to 0.01494. To validate pathology preservation, we compared lesion overlap, volume, intensity, and morphology. For downstream segmentation, nnU-Net performance improved from 0.6992 to 0.7137 Dice after augmentation with 50 synthetic cases, representing a modest but statistically significant improvement (paired t-test, p = 0.016). Tissue volume analysis showed no significant differences for CSF and GM compared to real images, while WM volume was lower in synthetic images. Conclusions: POWDR provides a framework for generating diverse, pathology-preserving synthetic MRI data. The results suggest potential utility for data augmentation while maintaining clinically relevant lesion characteristics.
Insights
We developed POWDR, a novel framework for generating synthetic 3D MRI data. This pathology-preserving outpainting method enhances medical imaging datasets, improving machine learning model performance for segmentation tasks.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Machine Learning
Background:
- Medical imaging datasets often face class imbalance and a scarcity of pathology-rich cases, hindering machine learning model performance.
- Existing methods struggle with generating realistic synthetic data that preserves crucial pathological details.
Purpose of the Study:
- To introduce POWDR, a pathology-preserving outpainting framework for 3D MRI data generation.
- To address limitations in medical imaging datasets by creating diverse, anatomically plausible synthetic data.
Main Methods:
- Leveraging wavelet-domain conditioning to enhance high-frequency details and reduce blurring in latent diffusion models.
- Implementing a random connected mask training strategy to improve data diversity and prevent conditioning-induced collapse.
- Evaluating POWDR on brain MRI (BraTS) and extending its application to knee MRI.
Main Results:
- Quantitative metrics (FID, MS-SSIM, LPIPS) confirmed the realism of generated images.
- Random connected mask training significantly improved data diversity, as indicated by reduced cosine similarity and increased KL divergence.
- Downstream segmentation tasks showed modest but statistically significant performance improvements (Dice score increase from 0.6992 to 0.7137) after augmenting with synthetic data.
- Synthetic images demonstrated comparable tissue volumes for CSF and GM, with a slight reduction in WM volume.
Conclusions:
- POWDR effectively generates diverse, pathology-preserving synthetic MRI data.
- The framework shows potential for data augmentation in medical imaging, maintaining clinically relevant lesion characteristics.
- This approach can help overcome data limitations in training machine learning models for various medical imaging tasks.
