Related Experiment Video
Updated: Aug 14, 2026
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
Tumor-SAM: Segment Anything Model for Semi-automatic Lung Tumor Segmentation in CT
1Medical Image Processing Group, Department of Radiology, University of Pennsylvania, Philadelphia, PA, 19104, United States.
Summary
This study introduces Tumor-SAM, an enhanced Segment Anything Model for semi-automatic lung tumor segmentation in CT scans. It improves accuracy and robustness over existing methods.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Accurate lung tumor segmentation in CT scans is crucial for radiomics and treatment assessment.
- Manual segmentation is labor-intensive and prone to inaccuracies.
- Existing automated methods struggle with tumor heterogeneity and ambiguous boundaries.
Purpose of the Study:
- To develop an improved Segment Anything Model (Tumor-SAM) for semi-automatic lung tumor segmentation.
- To enhance segmentation accuracy and robustness compared to current deep learning and traditional machine learning approaches.
Main Methods:
- Proposed Tumor-SAM integrates U-Net for multi-scale feature extraction and a novel ellipse prompt.
- The model first detects the lung region of interest (ROI) to minimize surrounding tissue interference.
- An ellipse prompt, defined by center, axes, and rotation, was designed for better tumor shape and location capture.
Main Results:
- Tumor-SAM achieved an average Dice index of 0.84±0.13.
- The average Hausdorff distance was 7.25±6.24 mm across 164 testing scans.
- Demonstrated good accuracy and robustness in lung tumor segmentation.
Conclusions:
- Tumor-SAM offers a promising semi-automatic solution for lung tumor segmentation.
- The U-Net integration and ellipse prompt contribute to improved segmentation performance.
- This method aids in more accurate radiomics analysis and treatment assessment.
