Diffusion model enhances kidney tumor images for better segmentation
Diffusion-Based Tumor Inpainting for Renal Segmentation under Clinical Data Scarcity
Computer Vision and Pattern Recognition
Summary
Medical AI programs struggle to find tumors in kidney scans because they don’t have many examples to learn from. The authors created a method that uses a type of AI called diffusion models to add realistic fake tumors to healthy kidney images, making it easier for AI to learn. They compared different ways to add these fake tumors in 2D, 2.5D, and 3D and found that 2.5D works just as well as 3D but uses less computing power. This helps reduce errors when the AI tries to find tumors, especially when real tumor examples are very few.
What this means in practice
- •For clinical imaging teams: Improve kidney tumor detection in CT scans using minimal annotated tumor data through enhanced image augmentation.
- •For medical ai developers: Incorporate 2.5D diffusion-based tumor inpainting to efficiently augment training datasets for renal tumor detection.
Authors
Ekaterina Sedykh, Salme Ussanov, Dmytro Fedorenko, Dmytro Fishman
Abstract
Deep learning segmentation of renal tumors requires large annotated datasets, yet clinical deployments typically offer only a handful of tumor-positive cases from the target site. We propose a diffusion-based inpainting framework that synthesizes anatomically plausible renal tumors within healthy CT scans, requiring no additional annotation, and provide the first systematic comparison of 2D, 2.5D, and full 3D (MAISI) synthesis strategies for this task. Training the diffusion model on public data (KiTS23, KIRC) and evaluating nnU-Net segmentation on a internal cohort across three low-data regimes, we find that 2.5D and 3D augmentation substantially reduce false positives (from $\sim$18--20\% to $\sim$3--6\%) while maintaining Dice, whereas 2D provides no consistent benefit. Crucially, the proposed 2.5D method matches full 3D synthesis on every metric at substantially lower computational cost, indicating that local volumetric consistency alone is sufficient for effective augmentation in data- and resource-scarce clinical settings.