AI detects Diabetes with a 10-second Voice Test

By: | October 26th, 2023

Image by Pixabay

Traditionally, finding prediabetes and Type 2 diabetes has meant expensive in-person tests, like blood work at a clinic or lab, often accompanied by extended wait times for results. These high costs and waiting periods often deter people from getting screened for these conditions.

High costs and waiting times could soon become a thing of the past

Recently, scientists found a method to determine if someone has diabetes by having them speak a few sentences on their smartphone.

Researchers at the international biotech company Klick Labs conducted a study that included 267 participants, comprising 192 non-diabetic individuals and 75 with type 2 diabetes diagnoses.

Each participant used a smartphone app to record a specific phrase up to six times a day for two weeks, resulting in recordings lasting six to ten seconds, depending on speaking speed.

Findings linked voice changes to type 2 diabetes development

Analysis of 14 acoustic characteristics from 18,465 recordings revealed consistent differences in features like pitch and intensity between diabetic and non-diabetic participants. Although these differences were inaudible to the human ear, identified by signal processing software.

The scientists developed an AI program that, when combined with patient data, achieved an 89% accuracy in identifying type 2 diabetic women and 86% accuracy for men. These numbers are expected to improve with further technology refinement.

In comparison, traditional fasting blood glucose tests were 85% accurate for both genders, while glycated haemoglobin and oral glucose tolerance tests scored 91% and 92% accuracy, respectively.

Nidhi Goyal

Nidhi is a gold medalist Post Graduate in Atmospheric and Oceanic Sciences.

More articles from Industry Tap...