Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study enhances diffusion generative models for synthetic medical image generation by introducing Fréchet Distance loss (FD-loss) to better capture the complex morphological characteristics of heterogeneous tumors. Traditional denoising objectives often smooth out critical structural details, but FD-loss aligns feature statistics in a pretrained encoder space, leading to improved fidelity in tumor representation. The integration of FD-loss across various architectures and datasets results in over 5% improvement in tumor Dice Similarity Coefficient (DSC) for downstream segmentation tasks, indicating its effectiveness in clinical applications.
Fréchet Distance loss boosts synthetic medical image quality, leading to over 5% improvement in tumor segmentation accuracy.
Diffusion generative models have demonstrated immense potential for synthetic medical image generation. However, these models often struggle to capture complex morphological characteristics of heterogeneous tumors with irregular boundaries, limiting their utility for downstream clinical tasks such as segmentation. This limitation stems from the standard denoising objective: minimizing a per-pixel error, which smooths high-variance irregular structures characteristic of tumors. To address this, we propose finetuning these generative models with Fr\'echet Distance loss (FD-loss). FD-loss aligns the first and second order feature statistics of real and generated images in a pretrained encoder space, encouraging the generator to capture complex structural variations characteristic of heterogeneous tumors. We integrate FD-loss across diverse architectural settings, using both natural- and medical-image encoders on multiple liver and brain cancer datasets spanning CT and MRI modalities. Downstream segmentation networks trained on our FD-regularized synthetic data consistently achieve superior performance, improving tumor DSC by $>$$5\%$ over unregularized synthetic augmentation alone. Qualitative analysis suggests these gains are associated with more faithful tumor synthesis and fewer segmentation hallucinations. Our results show FD-loss as an effective regularizer for medical image generative models to improve clinical workflows.