Artificial Intelligence (AI) has shown promising results in healthcare and medicine. Of late, another research conducted at the Ben-Gurion University of Negev developed an innovative technique of artificial intelligence to protect or safeguard medical devices from malicious threats and operating instructions during a cyberattack. It can even protect against the system and human errors.
At the International Conference on Artificial Intelligence (AIME) 2020, head of research, Tom Mahler, presented his research on August 26, 2020, “A Dual-Layer Architecture for the Protection of Medical Devices from Anomalous Instructions.”
A Ph.D. candidate, Mahler, under Professor Yuval Shahar and Professor Yuval Elovici of the BUG in the Department of Software and Information Systems Engineering, or SISE.
Complicated medical devices like Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) have PCs sending instructions to control the systems. Anomalous, or abnormal instructions, expose patients to harmful threats, like radiation overexposure, functionally manipulated medical images, or manipulated device components. Threats may also result due to human errors like configuration mistakes of technicians, cyberattacks, or PC software bugs.
Mahler, as an important part of his Ph.D. research, developed a unique technique with the help of artificial intelligence. The technique analyzes instructions sent from the computer to a physical component with the help of another innovative architecture to detect anomalous instructions.
Mahler expressed that the team has developed an innovative dual-layer architecture to prevent anomalous instructions. This architecture detects two key anomalous instructions – (i) context-free or CF type of anomalous instructions, and (ii) context-sensitive or CS type of anomalous instructions. The former comprises unlikely instructions or values like providing 100x radiation more than the typical, whereas the latter combines values and instruction parameters but are anomalous relative to specific context like mismatched type of scan, or mismatched age or weight of the patient, or any potential diagnosis. To specify, he said that a normal medical instruction given to an adult is potentially dangerous or anomalous for an infant. These type of instructions are only classified using the first layer of CF. However, when the second layer, or CS layer, is added, it can also be detected.
This new architecture was evaluated by the team of researchers in the CT (computed tomography) domain with 8,277 recorded instructions for CT and evaluated CF layer with 14 different types of unsupervised anomaly detection algorithms. Thereafter, they evaluate the layer of CS for different contexts of clinical objectives with five distinct supervised classification algorithms for context respectively.
When the second CS layer was added to the architecture, it improved anomaly detection performance from a score of 71.6 percent F1 using the CF layer, to somewhere between 82 percent and percent, depending on the part of the body or the clinical objective. Besides, this CS layer also detects CS anomalies with the help of semantics of the procedure of the device, a type of anomaly that the CF layer cannot detect.
Other members of the research team are Professor Yuval Shahar – Head of the Department of Medical Informatics Research Center at the BGU and Department of Software and Information Systems Engineering (SISE); Professor Yuval Elovici, Head of the Department Cyber Security Research Center of BGU, and a senior researcher from the Medical Informatics Research Center, Dr. Erez Shalom.
American Associates, the Ben Gurion University of the Negev, also called AABGU, has an important role in leading the vision of David Ben Gurion. Together, they wish to create an excellent educational institution and set out to research in the deserts of Israel, nurturing the incredible Negev community and sharing the expertise of the University locally and across the world. Some of the common activities are displaying academic excellence as well as cutting-edge research of the BGU through exceptional educational programs, informative communication, and events.
sepStream® is one of the most reliable medical imaging service providers. If you are looking for the best equipment and accurate imaging result, look no further. We use the latest algorithms to ensure patients get accurate results.