Search papers, labs, and topics across Lattice.
Volumetric Directional Diffusion (VDD) is introduced to address the fidelity-diversity trade-off in ambiguous medical image segmentation by anchoring the generative trajectory to a deterministic consensus prior. VDD iteratively predicts a 3D boundary residual field, restricting the generative search space and preserving anatomical coherence. Experiments on LIDC-IDRI, KiTS21, and ISBI 2015 datasets show VDD achieves state-of-the-art uncertainty quantification and competitive segmentation accuracy, providing clinicians with anatomically coherent uncertainty maps.
By guiding diffusion with a consensus prior, VDD generates anatomically plausible uncertainty maps for medical image segmentation, avoiding the hallucinations that plague standard generative approaches.
Equivocal 3D lesion segmentation exhibits high inter-observer variability. Conventional deterministic models ignore this aleatoric uncertainty, producing over-confident masks that obscure clinical risks. Conversely, while generative methods (e.g., standard diffusion) capture sample diversity, recovering complex topology from pure noise frequently leads to severe structural fractures and out-of-distribution anatomical hallucinations. To resolve this fidelity-diversity trade-off, we propose Volumetric Directional Diffusion (VDD). Unlike standard diffusion models that denoise isotropic Gaussian noise, VDD mathematically anchors the generative trajectory to a deterministic consensus prior. By restricting the generative search space to iteratively predict a 3D boundary residual field, VDD accurately explores the fine-grained geometric variations inherent in expert disagreements without risking topological collapse. Extensive validation on three multi-rater datasets (LIDC-IDRI, KiTS21, and ISBI 2015) demonstrates that VDD achieves state-of-the-art uncertainty quantification (significantly improving GED and CI) while remaining highly competitive in segmentation accuracy against deterministic upper bounds. Ultimately, VDD provides clinicians with anatomically coherent uncertainty maps, enabling safer decision-making and mitigating risks in downstream tasks (e.g., radiotherapy planning or surgical margin assessment).
