A Canadian application will soon make it possible, using AI, to screen for type 2 diabetes using a simple voice clip of 6 to 10 seconds. If effective, it would present a new, more accessible path (or voice) of detection.
Discovering possible diabetes just with your smartphone? The idea may seem simplistic, but it is actually being implemented in Toronto, where Klick Health, a scientific marketing company, is working on an AI engine capable of detecting disease based on the characteristics of your voice.
People affected by type 2 diabetes don’t always know it
Made by the pancreas, insulin helps absorb glucose and provides the body with the energy it needs to function. There are two basic types of diabetes:
- Type 1 diabetes, in which the pancreas has been attacked by the body’s immune system and is not able to produce insulin at all;
- Type 2 diabetes, which accounts for more than 90% of diabetes cases, where the body is not able to use insulin to break down glucose.
Diabetics suffer from poor blood circulation and are at acute risk of heart attack, stroke, amputation, blindness and kidney problems. If you are overweight, obese, or not physically active, you are at high risk of getting type 2 diabetes. Yet people who have it often don’t know it. Globally, 462 million people suffer from type 2 diabetes, but at least half are unaware of it.
The smartphone, more accessible than a laboratory test
Faced with this scourge, technology has not yet found a simple and affordable way to detect the presence of a common disease like diabetes. The most common test today is still one that measures blood sugar (called fasting blood glucose or FBG), which involves an overnight fast and a visit to a clinic, as does the increasingly popular test of glycated hemoglobin (A1C), which does not require fasting. Of course, they are not really “complicated” but can create a psychological or financial barrier for certain people to go and get tested.
It is in this niche that Klick Health has positioned itself with its idea of voice detection via smartphone: this would offer a fast, almost free and very accessible alternative since everyone has a phone today. All that remains is of course to have a detection product at the cutting edge of science.
AI dissects our voice like no other
However, here we are in the era of vocal biomarkers, where AI analyzes voice patterns and characteristics in many pathologies (Alzheimer’s, Parkinson’s, etc.), including diabetes.
The Klick company recorded the voices of 267 people who did not suffer from diabetes (79 women and 113 men) or who had already been diagnosed with type 2 diabetes (18 women and 57 men). Over a two-week period, participants recorded a short sentence: “Good morning, how are you doing ? What is my glucose level right now? — six times a day on their smartphone. This process generated over 18,000 voice samples, from which 14 acoustic features differing in prevalence or intensity across participants were selected.
“Diabetes tends to erode both nerves and muscles in men, impacting the robustness of their voice. On the other hand, women who have a higher correlation between depression or anxiety and diabetes tend to experience increased tone.” explains Jaycee Kaufman, one of the company’s scientists to ZDNET magazine.
Using these markers, Klick’s AI-powered model achieved surprising accuracy. The model test for diabetes was 89% accurate when evaluating women and 86% when evaluating men.
A first indication… To be confirmed with your doctor
As impressive as the method and the result are, the company itself believes that its technology should not supplant medicine. ”It is intended to be a first step towards a possible diagnosis of diabetes.”
All tests have false positives and other errors. It is therefore crucial for those who will use the test to reinforce their results with other conventional tests and validation from a professional. “However, given the pernicious role of diabetes in accelerating serious health problems, an early warning system in the form of a voice test, could save many lives and prompt others to seek early treatment .”
The reliability of this application will need to be confirmed in a larger population with different ethnic backgrounds (all participants were recruited in India) and different languages before it can be made available.