Computed tomography (CT) is currently one of the most widely used techniques to diagnose various diseases for its excellent performance and speed in obtaining images or cross-sections of bones, blood vessels, and soft tissues of the body. It has also become one of the radiological procedures that involves the highest exposure to radiation, equivalent to six months to five years of unprotected exposure to the sun.
To turn this around, Dr. Violeta Chang Camacho, an academic at the Usach Department of Computer Engineering, along with Dr. Aline Xavier, from the Biomedical Civil Engineering program, in collaboration with Dr. José Manuel Saavedra Rondo (project director), Dr. José Francisco Delpiano Costabal, and radiologist Héctor Henríquez Leighton, were awarded the Fondef IDeA I+D 2024 project “Reconstruction of low-dose radiation computed tomography using accelerated generative models based on diffusion techniques.”
Dr. Chang, who is the acting director of the project, explains that CT scans have a standard radiation dose that is used worldwide. However, if that dose is lowered, “the images of the lesions lose sharpness. If the radiation dose is lowered to 50%, the smallest lesions, measuring 1 to 3 centimeters, become invisible. Our goal is to use artificial intelligence methods to generate images from low-radiation images as if they were taken with a regular dose, without compromising diagnostic accuracy,” she says.
To achieve this, we propose developing artificial intelligence models based on generative prototypes, which will learn the distribution of CT scans with standard radiation doses and then reconstruct them from images with reduced doses. Diffusion models, which represent the latest in content generation, will be used, as well as an approach based on stochastic differential equations to streamline the inference process.
The results of this project will have a significant impact on the health of both medical staff and patients, especially children and young people, since “the effect of CT scans may not be apparent now, but it will be in thirty or forty years, when those pediatric patients reach adulthood,” she warns.
Once the accelerated generative models are obtained, they need to be tested in a real environment. This will be done by applying the innovation to imaging phantoms and patients at Clinica Santa María hospital, which has dual-source equipment, with the goal of eventually incorporating it into routine clinical practice.
Multidisciplinary research
The team developing this innovation in medical imaging includes professionals from the fields of biotechnology and radiology, as well as electrical and computer engineers and master's students from our university.
For Dr. Violeta Chang, having different disciplinary perspectives helps “reach a consensus on a path for developing new models in artificial intelligence with added value, which is what we are looking for in the field of computer science.” There are scientific articles that propose methods to solve this problem, but they do so from a technical perspective and stray from the clinical field. They never get to the point of assessing whether there is any impact on doctors or patients, which is where we want to add value,” she emphasizes.