Related Experiment Video
Updated: Jul 9, 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
Opportunistic Promptable Segmentation: Leveraging Routine Radiological Annotations to Guide 3D CT Lesion
Samuel Church1, Joshua D Warner2, Danyal Maqbool3
1Department of Computer Sciences, University of WI-Madison, Madison, WI, USA. sdchurch@wisc.edu.
Journal of Imaging Informatics in Medicine
|July 7, 2026
Summary
This study introduces SAM2CT, a novel AI model that converts simple radiologist annotations like arrows and lines into 3D segmentations for CT scans. This method efficiently generates valuable datasets for machine learning in medical imaging.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Machine Learning
Background:
- Developing machine learning models for CT imaging requires large, diverse, and annotated datasets.
- Manual 3D segmentation is time-consuming and costly, hindering dataset creation.
- Radiologists often provide sparse annotations (lines, arrows) stored in DICOM objects.
Purpose of the Study:
- To develop a method for converting sparse radiologist annotations into 3D segmentations for CT imaging.
- To introduce SAM2CT, the first promptable segmentation model for converting radiologist annotations into 3D CT segmentations.
- To enable the creation of large-scale annotated datasets from existing clinical data.
Main Methods:
- Proposed SAM2CT, a promptable segmentation model extending SAM2.
- Augmented prompt encoder to accept arrow and line inputs.
- Introduced a memory-conditioned module (MCM) for improved volumetric segmentation accuracy across slices.
Main Results:
- SAM2CT achieved high Dice Similarity Coefficients (DSC) on public benchmarks (0.649 for arrows, 0.757 for lines).
- Generated clinically acceptable 3D segmentations in 87% of cases using existing PACS annotations.
- Demonstrated strong zero-shot performance on emergency department findings like abscesses (DSC=0.610) and gallstones (DSC=0.725).
Conclusions:
- SAM2CT effectively converts sparse radiologist annotations into accurate 3D CT segmentations.
- Leveraging historical grayscale softcopy presentation state (GSPS) annotations is a scalable approach for generating large-scale 3D CT segmentation datasets.
- This opportunistic promptable segmentation paradigm significantly advances AI development in medical imaging.
