A Promptable 3D-CT Foundation Model-Based Approach for Pulmonary Embolism
An interactive AI tool segments blood clots in CT scans accurately and quickly, potentially helping clinicians assess stroke severity faster.
ClotIA, a fine-tuned SAM2-based foundation model, achieves DSC 0.83 for interactive 3D pulmonary embolism clot segmentation, outperforming both nnUNet and baseline performance, and enables rapid blood clot volume estimation that may support clinical severity assessment. The model represents a step toward clinical automation of a currently time-consuming manual task.
What the study was
- Study design
- Model development and validation study
- Population
- Pulmonary embolism patients from RSPECT dataset (2020)
- Sample size
- 309
- Category
- Diagnostics
- Maturity
- Exploratory
- Journal
- Cardiovascular and Interventional Radiology
Why it surfaced
Foundation model for PE segmentation with strong volume correlation; clinical PE severity assessment tool. Limited by single-dataset validation (RSPECT 2020). Interesting SAM2 fine-tuning approach.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.