AI That Saves Lives: The Chatbot That Can Detect A Heart Attack Using Machine LearningAdobe Stock
Cardiac Arrest—Biggest Killer on Earth
There are numerous out-of-hospital cardiac arrests annually—3,500 in Denmark, 30,000 in the UK and more than 350,000 in the United States. They are often deadly, responsible for 15 percent of all deaths in Western countries. Corti listens and learns from emergency calls with similar machine learning online courses technology that allows ad servers to understand an individual’s product preferences to then deliver appropriate ads. Instead of pushing products to consumers, Corti is able to serve human emergency operators information so they are better able to respond to an emergency call.
According to the American Heart Association, every second counts in a heart incident. Each minute that passes without beginning CPR leads to a 7 to 10 percent decline in survival rates.
The Corti hardware device, The Orb, (designed by Danish lamp designer Tom Rossau and Corti chief product offer Yuan Nielsen to be small, aesthetically appealing, organic and roughly the size of a Google Home speaker) was rolled out in Copenhagen this summer. As part of its training, Corti analyzed 161,000 emergency calls that originated in Copenhagen in 2014. There were 2,000 cardiac arrest calls in that year, and Corti accurately diagnosed 93 percent of those compared to 73 percent accuracy achieved by human operators. Not only did Corti accurately diagnose at a better rate, but it was also able to do so 30 seconds faster. This is an important difference when every minute matters.
How Corti Works in Real-Time on Emergency Calls
When a call is placed to an emergency call center in Copenhagen, a human operator responds while Corti listens in using speech recognition to understand the conversation. Similar to other machine learning online courses technology, Corti analyzes the words and other information such as background noise of the call in real-time to “learn” what signals a cardiac incident. It was not taught to identify anything in particular but has “taught” itself based on analyzing thousands of calls prior and continually improving as it gains more experience—the ultimate in on-the-job training.
If Corti determines factors that indicate a cardiac arrest, it signals the human operator who can advise the caller to begin CPR. Corti can identify what non-verbal sounds are important and sift through background noise such as sirens. Additionally, Corti can help human operators by double-checking other information on the call such as verifying the address information.
Corti’s Future: Human Assistant
After initial success in Copenhagen, the European Emergency Number Association (EENA) with members in 80 countries serving cities such as Munich, Milan, Paris and London, rolled out Corti to other cities for additional trials. There is international demand for the Corti device and service, but due to different privacy standards, details of how it will be deployed are still being worked out.
Since there is so much data available to learn from in the medical field, it is an ideal environment for artificial intelligence online courses applications. To expand its support to the emergency response community, Corti developers are currently working on intelligence to detect other ailments such as stroke, drug overdoses and more to assist human operators. There is other software development underway to add functionality that helps filter and flag information for further review.
Will Corti replace human operators? Corti CEO Andreas Cleve emphasizes that Corti is just a tool to supplement human intelligence and equip emergency operators with a resource to do their job more effectively. He told Fast Company, “These people are handling more or less the worst days of our lives but they have no tools to do it.” He explains that when it comes to health, people prefer human contact. If humans can do their jobs better with the support of AI, especially when it means the difference between life or death, Corti will likely be viewed as harmless as a Google search query.