Corti is an AI Equipped Digital Assistant That Can Do Medical Diagnoses Over the Telephone

June 20, 2018 – Corti SA, a Denmark-based software company, has created a program that analyzes conversations over the telephone to predict a diagnosis of a major heart attack. How accurate is it? According to the latest study done by the Danish National Institute of Public Health, Copenhagen EMS, and the University of Copenhagen, the software is accurate in 93% of cases versus 73% for human emergency medical dispatchers.

How is Corti able to be better than humans in diagnosing a major heart attack over a telephone line?

It is a voice-based interface to a neural network, a self-learning algorithm that mimics how the human brain learns. Expose the software to vast numbers of recorded emergency calls, to different dialects and accents, and even foreign languages and the algorithm with its speech recognition capability can determine the incidence of a major heart attack with high levels of accuracy.

Where emergency dispatchers use scripts and decision trees as they listen to callers to help determine a diagnosis, Corti hears more than the words and is not limited to the structured systems that human operators utilize. Corti picks up on non-verbal cues such as the breathing pattern of the caller. Corti also hears background noises and compares the received input with a vast store of data based on millions of historic emergency dispatch calls. As it understands the immediate incident it can then provide alerts and recommendations to the emergency dispatcher.

What are the results so far?

Corti is proving to be much faster than human dispatchers in making its heart attack diagnosis, an average of 48 seconds versus more than a minute and a quarter with just a human emergency operator on the line. And this is where seconds matter, the difference between surviving a heart attack or dying. If Corti tells the dispatcher it is certain in its diagnosis, it also can help find nearby resources like a portable defibrillator. It can advise the dispatcher to tell the caller to begin CPR.

In testing, the initial study had the software listen to 150,000 recorded emergency calls. The accuracy of diagnosis astounded the Copenhagen EMS service, so much so that they considered it a fluke and asked for a retest on an additional group of recorded calls. Once more Corti produced far more accurate predictions than human operators alone. Errors in diagnosis amounted to 2% of the total calls which matched the number in human operator predictions.

Since the spring Copenhagen EMS is using Corti in a randomized live call trial. Based on positive results from this current use of Corti, Copenhagen EMS is planning to roll the software out to all emergency hotline services.

Can Corti do more than diagnose major heart attacks?

The developers of Corti believe it can be used for many other types of emergency medical calls and presently have the software listening and learning through recorded conversations about mild heart attacks and strokes. It shouldn’t be long before Corti is collaborating with medical teams and emergency medical services to help save many more lives.