Coronary Angiography with High Radiation Exposure: AI Application Cuts Dose by More Than Half

Sihyuk Kang's Team at Seoul National University Bundang Hospital
Develops Frame Interpolation Model 'Angio-FILM'
Experts Unable to Distinguish AI Intervention

A technology has been developed that utilizes generative artificial intelligence (AI) to reduce radiation exposure from coronary angiography to less than half, while maintaining the same level of image quality as conventional methods.

From the left, Professor Sihyuk Kang, Cardiovascular Medicine, Seoul National University Bundang Hospital; Researcher Hui Kwon; Researcher Seyoung Park. Seoul National University Bundang Hospital

On July 13, Seoul National University Bundang Hospital announced that the research team led by Professor Sihyuk Kang from the Department of Cardiovascular Medicine (with Hwi Kwon and Seyoung Park as first authors) has developed a generative AI-based image interpolation model called 'Angio-FILM.' This model enables smooth imaging even with low-frame-rate coronary angiography, which reduces radiation dose to less than half of current standards.

Coronary angiography is a procedure in which a contrast agent is injected and the shape and blood flow of the heart's vessels are examined using continuous X-ray imaging. It is widely used for the detailed observation of blood vessels in the diagnosis and treatment of coronary artery diseases such as myocardial infarction. To match the rapid heartbeat of the heart and coronary arteries, images are taken at a rate of 10 to 15 frames per second. However, a higher frame rate also increases radiation exposure for both patients and medical staff.

If the frame rate is lowered to reduce radiation exposure, the time gap between scenes increases, which can cause interruptions or jitter in the movement of blood vessels. As a result, low-frame-rate imaging is difficult to use widely in clinical practice. Ultimately, patients are directly exposed to radiation during the procedure, while medical personnel repeatedly face the risk of radiation exposure throughout the day by wearing heavy protective equipment for precise treatment.

The Angio-FILM model developed by the research team is a solution that uses AI to generate intermediate frames expected to exist between recorded scenes, thereby reducing radiation exposure while maintaining conventional image quality. By recording at 7.5 frames per second—half the conventional frame rate—but using AI to interpolate the missing frames, the system restores image quality equivalent to 15 frames per second. As a result, it is estimated that radiation exposure can be reduced by more than half.

Schematic of the Angiography Frame Interpolation Model Angio-FILM. Seoul National University Bundang Hospital

The research team explained, "Since these images are also used for precision procedures, Angio-FILM is designed to reliably reproduce the rapid and nonlinear motion of the heart and coronary arteries." Instead of simply calculating the mean value between preceding and following frames, the team increased stability by separating the spatial and temporal analysis algorithms and computing the path using only the core elements of the image. This was accomplished by applying a 'Latent Flow Matching' technique.

In fact, in a 'Turing test' conducted by the research team, 30 medical specialists were asked to distinguish between 600 original and AI-interpolated images. Even when they were explicitly aware of AI involvement, the specialists' ability to identify the AI images did not significantly differ from random selection (50%). This demonstrates the high precision of the AI-generated images, which are indistinguishable from the originals even to trained experts. The difference in coronary lumen diameter between original and AI-interpolated images was only 0.18 mm, alleviating concerns about anatomical distortion.

Professor Sihyuk Kang stated, "Physical improvements to reduce radiation exposure during coronary angiography have reached their limits. If Angio-FILM, whose clinical reliability has been confirmed through this research, is adopted in practice, it could make a significant contribution to reducing radiation exposure for both patients and medical staff."

The results of this study were published in the latest issue of 'npj Digital Medicine,' a sister journal of Nature.

This content was produced with the assistance of AI translation services.