Prior studies have demonstrated that short voice recordings, when combined with AI technology, can diagnose various other illnesses, including COVID-19.
A new study suggests that individuals could potentially receive a diabetes diagnosis by providing a brief voice recording from their mobile phone. The researchers developed an AI model capable of identifying diabetes in a person with nearly 90% accuracy using a mere six to ten seconds of audio, combined with basic health information such as age, sex, height, and weight.
Brief Voice Recordings Can be Employed to Diagnose Diabetes in Individuals
Klick Labs enlisted 267 participants for the research, which included individuals who were already diagnosed with type two diabetes.
Each participant was requested to record a specific phrase six times daily for a duration of two weeks using their phone. The research team employed AI to examine over 18,000 voice samples in search of acoustic variations between those with diabetes and those without. These variations included pitch changes resulting from type two diabetes, which are imperceptible to the human ear.
The model achieved an accuracy rate of 89% for women and 86% for men.
Potential Revolution in Diabetes Screening
Jaycee Kaufman, the author of the study, indicated that the findings have the potential to revolutionize diabetes screening methods.
In the UK, over 90% of adults with diabetes are affected by type two, but many remain unaware for years due to the general or absent symptoms.
Traditionally, individuals need to visit their GP and undergo urine and blood tests for diabetes screening.
Ms. Kaufman stated, “Current detection methods often demand significant time, travel, and expenses. Voice technology has the potential to eliminate these obstacles completely.
Prior studies have indicated that voice recordings, when paired with AI technology, can serve as a diagnostic tool for various illnesses, including COVID-19.
Klick Labs anticipates that this technology might also have the potential to diagnose conditions such as prediabetes and hypertension.
This peer-reviewed study is now accessible in the Mayo Clinic Proceedings journal.