This system allows an automatic diagnosis of this condition in hospitals
12 July 2023
Coronary artery tortuosity is a rare condition where the arteries appear to "twist" and is a risk factor for heart attack or aneurysm. To detect it, when a patient presents with symptoms of coronary artery disease, coronary angiography (a diagnostic test where a contrast dye is used to study blood vessels that are not visible with conventional X-rays) is performed, however, detecting arterial tortuosity requires a thorough and time-consuming examination by cardiologists.
Therefore, a research team from the Institute of Physics of Cantabria (IFCA), a joint centre of the CSIC and the University of Cantabria, formed by Miriam Cobo, Lara Lloret, and Ignacio Heredia, with the collaboration of the Catholic University of Maule, the Autonomous University of Chile, the Regional Hospital of Talca and the University of Oviedo, have developed an artificial intelligence model based on deep learning, capable of detecting coronary tortuosity faster and with the same accuracy as a specialist. The work has been published in the scientific journal Scientific Reports (Nature).
Miriam Cobo has developed, together with Lara Lloret and Ignacio Heredia, this deep learning model. / IFCA.
"Our goal is to analyse the tortuosity of coronary arteries in coronary angiography with artificial intelligence techniques, to develop an algorithm capable of automatically detecting this condition in patients", explains Miriam Cobo, a researcher in the IFCA's Computation and e-Science Group.
To do so, the research team has used deep learning techniques, specifically the so-called convolutional neural networks, to classify patients as tortuous or non-tortuous, depending on their coronary angiography.
The developed model has been trained for two years with 658 coronary angiographies of 401 patients selected from a database of 18,000 people, and who attended between 2016 and 2022 to the Hospital de Talca (Chile), with symptoms of suffering from coronary heart disease.
During the study, these trained deep learning neural networks were found to have a sensitivity and accuracy comparable to the visual examination performed by cardiologists when detecting coronary artery tortuosity in visits. In addition, the deep learning model only requires a single image to perform the tortuosity detection, while the specialist examines the entire video sequence to make the diagnosis. Thus, the research has succeeded in developing, through imaging, an image-based learning model that improves diagnostic efficiency.
Examples of left coronary angiographies of patients with coronary artery tortuosity (a-b-c) and without coronary tortuosity (d-e-f). / Nature.
Artificial intelligence, a support for hospitals
Medical imaging already plays a key role in today's medicine for monitoring, diagnosis and evaluation of various personalised treatments. And artificial intelligence in this field can support, through convolutional neural networks, the classification and detection of diseases more quickly. "We believe that this model can serve as a first screening to predict the probability of a patient being diagnosed with coronary tortuosity, offering extra help to specialist cardiologists. In addition, we believe that this study could help to further validate future applications of AI techniques in cardiology," says Cobo.
For the researcher, this deep learning system has a promising future, and not only in the field of cardiology, "we will soon apply it to the study of images of patients with lung cancer", she announces.
Rebeca García / IFCA Communication
