Currently, if someone wants to see if they have type 2 diabetes, they have to travel to a clinic for blood tests and then wait for results. According to a recent study, however, a 10-second smartphone voice recording may soon deliver on-the-spot results immediately.
The study was conducted by scientists from international biotech firm Klick Labs, and involved 267 test subjects who had already been diagnosed as being either non-diabetic (192 people) or type 2 diabetic (75 people).
Each person was asked to record a specific spoken phrase on their own smartphone via an app, up to six times a day for two weeks. Depending on the speed at which each individual spoke, those recordings were six to 10 seconds long.
When 14 acoustic features of the resulting 18,465 recordings were analyzed, it was found that several of those features – such as pitch and intensity – differed in a consistent manner between the diabetic and non-diabetic participants. Although these differences couldn't be detected by the human ear, they could be picked up by signal processing software.
This finding suggests that developing type 2 diabetes causes subtle changes in a person's voice.
With that theory in mind, the scientists created an AI-based program that analyzes voice recordings along with patient information such as age, sex, height and weight. When tested on the volunteers, that program proved to be 89% accurate at identifying type 2 diabetic women and 86% accurate at spotting diabetic men.
Those numbers should improve as the technology is refined. For reference, the team found that traditional fasting blood glucose tests were 85% accurate for both sexes, while glycated hemoglobin and oral glucose tolerance tests were 91% and 92% accurate, respectively.
Plans now call for further voice tests to be conducted on a larger, more diverse population.
"Current methods of [diabetes] detection can require a lot of time, travel and cost," said Klick Labs research scientist Jaycee Kaufman, first author of paper on the study. "Voice technology has the potential to remove these barriers entirely."
The paper was recently published in the journal Mayo Clinic Proceedings: Digital Health.
Source: Klick Labs
