TECHNOLOGIST: How does your device work?
Michael Reibel Boesen: We developed a small instrument that can be placed on an emergency-service telephone operator’s desk to monitor conversations. It analyses what’s said and displays patient symptoms on a screen to help the operator come up with a diagnosis. Our technology relies on deep learning, which enables machines to grasp certain tasks through artificial neural networks.
T. What applications exist for this technology?
MRB. For now, we’ll be concentrating on health. Our device can pick up emergency calls involving heart attacks in 90% of cases – in other words, much more often than with human analysis. We’re thinking of extending its use to strokes, which are often difficult to diagnose by phone, as the symptoms are varied and unexpected. In the future, we also hope to evaluate whether the person at the other end is panicked or in pain.
T. What are your main challenges?
MRB. It’s not always easy to convince emergency services that a machine can be more efficient than their highly-trained personnel. To make our algorithm even more efficient, we have to feed it large quantities of data. These data, however, are rarely of high quality. Calls made to emergency services receive interference from surrounding noise like conversations or traffic.