Related Experiment Video
Updated: Jun 25, 2026
13:07
Optical Coherence Tomography Based Biomechanical Fluid-Structure Interaction Analysis of Coronary Atherosclerosis Progression
Published on: January 15, 2022
Video frame interpolation for coronary angiography using latent flow matching
Hwi Kwon1, Seyeong Park1, Do-Hyun Kim1,2
1Cardiovascular Center, Seoul National University Bundang Hospital, Seongnam-si, Korea.
NPJ Digital Medicine
|June 23, 2026
Summary
Angio-FILM, a generative AI model, synthesizes high-resolution coronary angiography videos from low-frame-rate inputs. This AI tool enhances temporal resolution while maintaining anatomical accuracy, aiding clinical applications.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Cardiovascular Medicine
Background:
- Coronary angiography involves inherent radiation exposure.
- Reducing frame rates lowers radiation but impairs temporal resolution.
Purpose of the Study:
- Introduce Angio-FILM, a generative AI model for coronary angiography video frame interpolation.
- Synthesize high-temporal-resolution videos from low-frame-rate inputs to reduce radiation exposure.
Main Methods:
- Developed Angio-FILM using a latent flow matching model.
- Trained the model on over 350,000 coronary angiography videos.
- Validated using internal, external, and open-source datasets.
Main Results:
- Qualitative expert evaluation showed Angio-FILM outperformed existing methods.
- Visual Turing tests indicated generated videos were indistinguishable from real angiograms.
- Quantitative analysis confirmed high anatomical fidelity with minimal deviation in measurements.
Conclusions:
- Angio-FILM effectively interpolates frames to create high-temporal-resolution coronary angiography videos.
- The AI model shows potential for reducing radiation exposure without significant loss of diagnostic information.
- Angio-FILM represents a significant advancement in applying generative AI to clinical practice.
