Uncertainty-Guided Coarse-to-Fine Tumor Segmentation with Anatomy-Aware Post-Processing

2025-04-16

AI Summary

To address tumor segmentation challenges in thoracoabdominal CT—including ambiguous boundaries, severe class imbalance, and anatomical variability—this paper proposes an uncertainty-guided coarse-to-fine two-stage framework. Stage I performs full-convolutional coarse localization; Stage II conducts anatomy-aware ROI selection by integrating lung overlap ratio, surface proximity, and component size constraints, followed by fine-grained segmentation driven by a dropout-based Monte Carlo uncertainty-weighted loss. The key innovation lies in the first integration of predictive uncertainty modeling with multi-scale pulmonary anatomical priors within a cascaded pipeline, jointly optimizing boundary refinement and false-positive suppression. Built upon the Swin UNETR backbone, our method achieves a 17.6% Dice score improvement (0.4690 → 0.6447) on the Orlando dataset, significantly reduces Hausdorff distance, lowers false-positive rate, and yields spatially more interpretable segmentations.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Segmentation
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Reliable tumor segmentation in thoracic computed tomography (CT) remains challenging due to boundary ambiguity, class imbalance, and anatomical variability. We propose an uncertainty-guided, coarse-to-fine segmentation framework that combines full-volume tumor localization with refined region-of-interest (ROI) segmentation, enhanced by anatomically aware post-processing. The first-stage model generates a coarse prediction, followed by anatomically informed filtering based on lung overlap, proximity to lung surfaces, and component size. The resulting ROIs are segmented by a second-stage model trained with uncertainty-aware loss functions to improve accuracy and boundary calibration in ambiguous regions. Experiments on private and public datasets demonstrate improvements in Dice and Hausdorff scores, with fewer false positives and enhanced spatial interpretability. These results highlight the value of combining uncertainty modeling and anatomical priors in cascaded segmentation pipelines for robust and clinically meaningful tumor delineation. On the Orlando dataset, our framework improved Swin UNETR Dice from 0.4690 to 0.6447. Reduction in spurious components was strongly correlated with segmentation gains, underscoring the value of anatomically informed post-processing.

Problem

• Improving tumor segmentation in thoracic CT scans
• Addressing boundary ambiguity and class imbalance
• Reducing false positives with anatomical post-processing

Innovation

• Uncertainty-guided coarse-to-fine segmentation framework
• Anatomically aware post-processing with filtering
• Cascaded models with uncertainty-aware loss functions