Researchers at Seoul National University Bundang Hospital, led by Professor Sihyuk Kang, have developed Angio-FILM, a generative AI–based image interpolation model for coronary angiography.

Angio-FILM enables imaging at a low frame rate of 7.5 frames per second by interpolating intermediate frames to achieve perceptual quality equivalent to 15 frames per second, thereby halving radiation exposure.

The study found a maximal lumen diameter difference of just 0.18 mm between original and AI-interpolated images, indicating minimal anatomical distortion.

Results were published in npj Digital Medicine, a Nature sister journal, providing peer‑reviewed validation of Angio-FILM.

In a Turing-test–style evaluation with 30 medical specialists, AI-interpolated images were indistinguishable from originals, with 50% accuracy—consistent with random guessing—demonstrating high perceptual similarity.

Angio-FILM uses Latent Flow Matching to separate spatial and temporal analysis, preserving rapid, nonlinear heart and artery motion rather than simply averaging frames.

The researchers note that adopting Angio-FILM in clinical practice could substantially reduce radiation exposure for patients and healthcare staff.