StructSAM: structure-aware prompt adaptation for robust lung cancer lesion segmentation in CT
-
By
-
Mengjie Liu
-
Yuxin Yao
-
Jinyong Jia
-
Jiali Yao
-
Zhengze Huang
-
Ziyang Zeng
-
Guangjin Pu
-
Yan Wu
-
Yuqi Bai
-
Bin Wang
-
Lili Jiang
-
February 3, 2026
-
Clinical Scorecard: StructSAM: Adapting Prompts with Structural Awareness for Enhanced Segmentation of Lung Cancer Lesions in CT Imaging
At a Glance
| Category | Detail |
|---|
| Condition | Lung cancer lesion segmentation in CT imaging |
| Key Mechanisms | Structure-aware prompt adaptation integrating anatomical priors and 3D volumetric context into foundation model (SAM) for improved lesion delineation |
| Target Population | Patients undergoing CT imaging for lung cancer diagnosis and management |
| Care Setting | Oncology imaging and radiology departments utilizing CT for lung cancer evaluation |
Key Highlights
- StructSAM integrates shape- and topology-based anatomical priors to stabilize segmentation at low-contrast lesion boundaries.
- A lightweight 3D-aware adapter aggregates inter-slice contextual information ensuring volumetric continuity in CT scans.
- Domain-aware parameter-efficient fine-tuning enables robust generalization across datasets and institutions.
Guideline-Based Recommendations
Diagnosis
- Utilize CT imaging for accurate delineation of lung lesions to support early detection and precise staging.
- Incorporate volumetric segmentation methods that maintain inter-slice consistency for reliable lesion representation.
Management
- Apply structure-aware segmentation frameworks like StructSAM to improve lesion margin accuracy for treatment planning.
- Leverage domain-adaptive models to enhance robustness across different clinical cohorts and scanner protocols.
Monitoring & Follow-up
- Use volumetrically consistent segmentation outputs to assess longitudinal tumor response and growth patterns.
- Employ automated segmentation tools with anatomical priors to reduce variability in follow-up imaging assessments.
Risks
- Be aware that conventional 2D segmentation models may produce inconsistent lesion boundaries and volumetric discontinuities.
- Recognize that low-contrast lesions and ambiguous boundaries can lead to under- or over-segmentation without structural guidance.
Patient & Prescribing Data
Patients with lung cancer undergoing CT imaging for lesion assessment
Enhanced segmentation accuracy and volumetric consistency support precise tumor delineation, facilitating personalized treatment planning and monitoring.
Clinical Best Practices
- Incorporate anatomical priors such as organ masks and vesselness cues to guide segmentation models in ambiguous regions.
- Adopt 3D-aware adapters to capture inter-slice context and ensure volumetric coherence in lesion delineation.
- Utilize domain-aware fine-tuning strategies to maintain model performance across diverse clinical settings and imaging protocols.
References
