An open-source, automated machine learning approach for large-scale image retrieval for thoracic aorta analysis
Brian C Ayers1, Aaron D Aguirre2,3, Thoralf M Sundt1
1Division of Cardiac Surgery, Massachusetts General Hospital, Boston, MA 02114, United States.
JAMIA Open
|July 15, 2025
Summary
An automated pipeline accurately identifies thoracic aortic computed tomography (CT) scans for research. This tool enhances image retrieval from large CT scan databases, improving data selection for studies.
Area of Science:
- Medical imaging analysis
- Radiology informatics
Background:
- Thoracic aortic computed tomography (CT) scans are crucial for cardiovascular research.
- Identifying specific scan series from large, diverse databases presents a significant challenge.
Purpose of the Study:
- To develop an automated image retrieval pipeline for identifying thoracic aortic CT scan series.
- To accurately select scans showing the entire thoracic aorta with arterial phase contrast.
Main Methods:
- Developed an automated image analysis pipeline.
- Applied the pipeline to a cohort of 4184 chest CT scans.
- Utilized criteria for selecting entire thoracic aorta with arterial phase contrast.
Main Results:
- The pipeline identified 3435 (82%) relevant studies.
- Manual review confirmed high accuracy: 99.1% of selected scans were accurate, and 93.6% of excluded scans were appropriately excluded.
Conclusions:
- An open-source pipeline accurately retrieves specific aortic imaging studies from heterogeneous collections.
- The framework is adaptable for various clinical use cases and multi-institutional research.
