Canadian medical researchers trained artificial intelligence to accurately predict type 2 diabetes within just six to 10 seconds of hearing a patient's spoken voice.
Working with faculty at Ontario Tech University in Canada, Klick Labs scientists trained the AI using recordings from 267 people recruited from India.
Participants were asked to record a phrase on their mobile phones six times a day for two weeks.
From 18,000 individual recordings, the scientists focused on 14 vocal features in search of consistent, replicable differences between groups with and without type 2 diabetes.
Four of these audio features proved to be the most useful in accurately predicting who has diabetes and who does not.
The AI focused on a range of vocal characteristics, including subtle changes in pitch and intensity, and linked that data to basic health information, including the patient's age, gender, height and weight.
The researchers found that gender proved to be a decisive factor: AI was able to diagnose the disease with an accuracy of 89% in women, but with a slightly lower accuracy for men, reaching 86%.
“Our research highlights significant vocal differences between individuals with and without type 2 diabetes,” said Jaycee Kaufman, the paper's first author and a research scientist at Klick Labs, which plans to commercialize the program.
In the past, expensive personal diagnostic tests, including a blood test, were needed to screen for prediabetes and type 2 diabetes.
Among the most common tests used are the glycated hemoglobin (A1C) test, the fasting blood glucose (FBG) test, and the oral glucose tolerance tests (OGTT).
“Current detection methods can be time and cost intensive,” Kaufman noted in a statement accompanying the new study, published in Mayo Clinic Proceedings Digital Health. “Acoustic technology has the potential to completely remove these barriers.”
Source: Publication date: 19/10/2023- Daily Mail -https://r.rtarabic.com/w9f4