Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study addresses the limitations of single-slice propagation methods in 3D medical image segmentation by introducing a novel approach that utilizes three orthogonal seeds—axial, coronal, and sagittal—to enhance label propagation through volumetric data. By focusing on inference-time seed geometry rather than the training paradigm, the authors demonstrate that orthogonal seeding significantly improves segmentation metrics, achieving a 21.9% increase in Dice score and a 53.5% reduction in Average Hausdorff Distance compared to traditional single-axis methods. These findings highlight the importance of leveraging 3D anatomical information to optimize segmentation accuracy in clinical applications.
Orthogonal seeding can boost 3D organ segmentation performance by over 20% without the need for extensive training adjustments.
Dense voxel-level annotation remains a major bottleneck in 3D medical image segmentation. Single-slice propagation methods such as Sli2Vol reduce this burden by propagating one annotated seed slice through a volume using label-free registration. However, axial-only propagation accumulates errors with distance from the seed, especially in surface-distance metrics, because it ignores coronal and sagittal evidence and therefore underuses the 3D information available in CT/MRI volumes. To better leverage volumetric geometry, we study how key training and inference choices affect slice-propagation models, including single-axis versus multi-axis label-free registration, single-seed versus multi-seed propagation, and orthogonal seed configurations. Instead of propagating from a single axial seed, we use three orthogonal seeds---one axial, one coronal, and one sagittal---and fuse their propagated labels with a simple label-free rule. Our results show that the training paradigm has limited impact: an axially trained network applied to off-axis seeds captures nearly all the improvement, while explicit three-axis training adds little. Instead, performance is driven by inference-time seed geometry, especially orthogonality rather than the number of annotated slices, as a budget-matched three-axial control provides no benefit and can even degrade performance. On a multi-organ CT cohort, orthogonal seeding with the axial Sli2Vol backbone improves Dice by 21.9%, Normalized Surface Dice by 25.5%, and reduces Average Hausdorff Distance by 53.5% over the single-axis baseline.