Content
Pepole
CV
A group of researchers from Amirkabir University of Technology presented a solution based on machine learning and artificial intelligence to increase the accuracy in displaying coronary arteries.
| Last update: 2026/02/3 |
Akhondi, referring to the reasons for choosing this research, explained: In cardiac angiography, the interference of shadows from chest tissues and bones, as well as respiratory movements and heartbeat, makes the accurate diagnosis of coronary arteries difficult. This issue leads to the need for injecting more contrast agent and receiving a higher radiation dose by the patient.
She added: Therefore, the goal of this research was to solve this problem and provide a more accurate and safer method to improve the quality of blockage diagnosis by a specialist physician.
This Amirkabir University of Technology graduate, emphasizing that the results of this research have improved image clarity and increased accuracy in displaying coronary arteries, stated: This research has specifically taken an effective step towards advancing medical image processing technology and localizing novel neural network methods; because this project, by reducing processing time and increasing accuracy compared to other traditional algorithms, provides the possibility of better utilization in applied fields.
Akhondi, stating that the first phase of this research started with analyzing the shortcomings and existing methods and reviewing articles, continued: Then, two models based on neural networks were designed and implemented. In the first method, an improved SpyNet network was used for motion detection and Deep-Fillv2 was used for reconstructing removed areas. In the second method, a hybrid deep autoencoder architecture was utilized to extract background information from images before contrast agent injection and use it for reconstructing areas containing contrast agent.
According to her, at the end, an improved version of the second model was presented and compared with standard methods.
She mentioned lack of data as one of the challenges of this project and noted: In addition, training deep neural networks required time and very powerful systems with strong graphics cards. We managed to overcome these problems as much as possible.
According to this researcher, the results of this research can be used in the medical equipment and software industry, image analysis, and AI-based diagnostic assistance systems. However, on the other hand, due to the import nature of most devices, adding this software to the entire system is not easy, and currently, it can assist the physician's diagnosis alongside the device as an aid.
Akhondi considered the publication of two articles as other achievements of this thesis and added: The first article has been published in an international Q1-ranked journal, and the second article has been published in an Iranian biomedical journal.
She mentioned the innovative part of the project as including the use of a guided autoencoder network for reconstructing the background of images containing contrast agent from images without contrast agent, and added: Also, in designing deep neural networks, suitability for angiography images due to their grayscale and vessel delicacy has been considered in the design.
Akhondi listed the features of this project as including increasing the clarity of coronary arteries in angiography images, reducing the need to inject a large amount of contrast agent and consequently reducing associated risks for the patient, and added: Also, compared to traditional methods, it significantly reduces image processing time and can perform correction and processing even with having a single image containing contrast agent. Finally, utilizing deep neu
ral network technology enables intelligent extraction and reconstruction of the background and better separation of vessels, placing the project scientifically and applicably at the level of novel medical image processing methods.
She emphasized: Foreign examples such as methods based on RPCA or MAMR exist, but the current method is domestic and has faster performance. In this project, by eliminating the need for multi-temporal images, the radiation dose received by the patient is reduced, and the imaging process has become simpler and safer. This method, due to relying on advanced algorithms and deep learning, processes images faster and automatically while providing higher accuracy in separating vessels and background.
She emphasized: Foreign examples such as methods based on RPCA or MAMR exist, but the current method is domestic and has faster performance. In this project, by eliminating the need for multi-temporal images, the radiation dose received by the patient is reduced, and the imaging process has become simpler and safer. This method, due to relying on advanced algorithms and deep learning, processes images faster and automatically while providing higher accuracy in separating vessels and background.
According to the public relations report of Amirkabir University of Technology, Fatemeh Akhondi, a graduate of the university's Faculty of Biomedical Engineering, under the guidance and supervision of Dr. Farshad Almas Ganj, a faculty member of the university, implemented a research project titled "Background Removal of Angiography Images Using Deep Learning."
Related
View: 285 Time(s) |
Print: 25 Time(s) |
Email: 0 Time(s) |
0 Comment(s)