Computer Vision News - April 2023

39 Miguel Bernabeu cardiologists interpret an ECG. We’re using computer vision to extract information from them, and with the help of these algorithms, we can begin to standardize the process. ” ECG interpretation requires experience, but an expert may not always be available in every hospital at any given time. These algorithms can bridge that gap; furthermore, automated tools can help streamline the work of cardiologists and enable them to see more patients. The group is collaborating with the NHS as part of this project to gain insight into the requirements for integrating its algorithms into clinical trial systems. The Computational Medicine Group is also workingwith Aidence , a Dutch AI company, on a joint evaluation project called INPACT . Aidence has a commercial computer vision for people to work on. However, Miguel says the way forward is to develop more advanced methods of accessing highly sensitive information and handling it with the utmost care. “ Health Data Research UK allows researchers to access data that was not necessarily consented for research, ” he points out. “ No one has the legal right to release this data out in the open, but they provide sophisticated IT infrastructures for you to do your work in a way that ensures it remains confidential and is treated according to the regulations. I think the future will be in creating these infrastructures for us to deal with as much patient data as is legally allowed. ” As well as discovery science, the group has activities closer to implementation in practice. Currently, it is working on deploying computer vision algorithms for interpreting electrocardiograms (ECGs) in hospitals, supporting cardiologists to identify features in the ECG that indicate someone is presenting with a heart attack. “ When someone walks through the door at A&E with chest pain, it might be a heart attack or something else, ” Miguel explains. “ ECGs are one tool that gets used for diagnosis, and there tends to be quite a lot of variability between how different