CoreFormer: High-Precision Pulmonary Nodule Segmentation via Structural Core Anchors
Overview
CoreFormer is a novel segmentation framework that significantly improves pulmonary nodule delineation in chest CT scans by leveraging structural core anchoring and geodesic shape decoding. Evaluated on four public datasets, it achieves state-of-the-art boundary accuracy and topological consistency, outperforming existing methods.
Background
Accurate segmentation of pulmonary nodules is critical for early lung cancer diagnosis and treatment planning. Traditional voxel-wise methods often produce fragmented masks and fail to preserve anatomical topology due to low contrast and imaging noise. Recent implicit representation approaches improve smoothness but lack anatomical context, limiting their effectiveness. CoreFormer addresses these challenges by modeling nodules based on their intrinsic structural cores and anatomy-aware geodesic paths, enhancing segmentation fidelity.
Data Highlights
Dataset
Metrics Improved
Performance Highlights
LIDC-IDRI
Dice, Hausdorff distance, Boundary precision
Significant improvement over baselines
LNDb
Dice, Hausdorff distance, Boundary precision
Consistent state-of-the-art results
Tianchi-Lung (MosMedData)
Dice, Hausdorff distance, Boundary precision
Robust segmentation of complex nodules
NSCLC-Radiomics
Dice, Hausdorff distance, Boundary precision
High-fidelity topological segmentation
Key Findings
CoreFormer identifies the intrinsic topological core of pulmonary nodules, providing a stable structural anchor for segmentation.
It employs anatomy-aware geodesic paths to generate continuous and topologically consistent nodule boundaries.
The dual-branch decoder architecture synergistically combines structural core prediction with context-aware shape decoding.
Feature Manifold Regularization enhances discriminative feature learning, improving robustness against imaging noise and anatomical variability.
Extensive evaluation on four public datasets demonstrates superior performance in boundary accuracy and topological fidelity compared to existing methods.
Clinical Implications
CoreFormer’s high-precision segmentation can improve early lung cancer detection by providing more accurate and reproducible pulmonary nodule delineations. This facilitates better clinical decision-making for biopsy, follow-up imaging, and treatment planning, potentially reducing variability and workload associated with manual segmentation. Its robustness across diverse datasets suggests broad applicability in clinical practice.
Conclusion
CoreFormer represents a significant advancement in pulmonary nodule segmentation by integrating structural core anchoring with anatomy-aware geodesic decoding, achieving superior accuracy and topological consistency. This framework holds promise for enhancing automated lung cancer screening and management.
References
CoreFormer Study -- CoreFormer: A High-Precision Framework for Segmenting Pulmonary Nodules Using Structural Core Anchors and Geodesic Shape Representation
