Synthetic peak breast MRI images generated with segmentation guidance
Anguinus Sculpturae: Compositional Synthesis of Peak-Enhancement Breast DCE-MRI Scans
Computer Vision and Pattern Recognition
Summary
Dynamic contrast-enhanced breast MRI scans help detect cancer but require injecting a contrast agent, which can cause safety concerns and extra costs. To avoid this, the authors developed a way to create the peak-enhancement images from earlier scans without the agent, by predicting just the difference that contrast makes. Their method uses image segmentations to guide the process, ensuring it captures both the whole breast and the lesion areas accurately in one step. This approach performed best in tests on a standardized breast MRI challenge dataset.
What this means in practice
- •For medical imaging software developers: Create MRI imaging tools that produce synthetic contrast images to reduce the need for actual contrast agents in breast MRI scans.
- •For radiology hardware vendors: Integrate synthetic contrast generation algorithms into breast MRI systems to enhance imaging without additional contrast injections.$Commercial implications: Enables MRI manufacturers to offer safer and cost-effective breast imaging features, enhancing product appeal in healthcare markets.
Authors
Benjamin Hamm, Nico Albert Disch, Maximilian Rokuss, Yannick Kirchhoff, Constantin Ulrich, Klaus Maier-Hein
Abstract
Dynamic contrast-enhanced breast MRI (DCE-MRI) is rich in anatomical and perfusion information, but its reliance on gadolinium-based contrast agents raises safety concerns and adds cost. Virtual contrast enhancement, synthesizing post-contrast from pre-contrast images, is a promising alternative. We address the MAMA-SYNTH challenge task of predicting peak-enhancement breast MRI. Rather than adopting the full machinery of diffusion or flow matching, we observe that under a rectified, straight-line path the generative process collapses to a single difference prediction: the synthetic peak image is the pre-contrast image plus a predicted enhancement map, recovered in one forward pass. Around this we build Anguinus Sculpturae, a compositional pipeline in which nnU-Net segmentations of lesion, foreground and breast region guide two generators - one optimized for global fidelity, one for lesion structure through an asymmetric Tversky term routed via a frozen segmenter - composited region-wise with Gaussian-weighted blending. On the held-out Duke subset of MAMA-MIA our model achieves the best FRD and Dice among all evaluated variants, showing that single-step difference prediction with segmentation guidance suffices to recover both global fidelity and lesion structure. Code is available at https://github.com/MIC-DKFZ/AnguinusSculpturae.