0
0
Madrid, September 9-13, 2024 — A groundbreaking study presented at the Annual Meeting of the European Association for the Study of Diabetes (EASD) has unveiled a new artificial intelligence (AI) model capable of detecting type 2 diabetes (T2D) by analyzing voice recordings. The research, conducted by a team from the Luxembourg Institute of Health, demonstrates the potential of voice analysis to identify previously undiagnosed cases of T2D, achieving an accuracy of 66% in women and 71% in men.
The Promise of Voice Analysis
As many as 240 million adults globally are unaware they have diabetes, predominantly T2D, due to vague or absent symptoms. Early detection and treatment are critical in preventing severe health complications. The study’s lead author, Abir Elbeji, emphasized the limitations of current screening methods, which are often time-consuming, invasive, and costly. “Combining AI with voice technology has the potential to make testing more accessible by removing these obstacles,” she stated.
The study aimed to develop a voice-based AI algorithm that could effectively distinguish between individuals with and without T2D. Participants provided voice recordings of themselves reading specific sentences using their smartphones or laptops, contributing to a pool of data that included basic health information such as age, sex, body mass index (BMI), and hypertension status.
Study Findings and Methodology
A total of 607 adults participated in the Colive Voice study, with recordings analyzed for various vocal characteristics, including pitch, intensity, and tone. Two advanced techniques were utilized: one capturing up to 6,000 detailed vocal features, and another employing a deep-learning approach that focused on 1,024 key vocal traits.
The results indicated that the AI model performed effectively, correctly identifying 71% of male and 66% of female T2D cases. Notably, the algorithm showed enhanced accuracy in females aged 60 and older and individuals with hypertension. Additionally, there was a 93% agreement with the questionnaire-based American Diabetes Association (ADA) risk score, indicating comparable performance between the voice analysis method and a recognized screening tool.
Next Steps in Research
While the findings are promising, co-author Dr. Guy Fagherazzi cautioned that further research and validation are essential before this approach can be widely adopted as a first-line screening strategy. “Our next steps are to specifically target early-stage type 2 diabetes cases and pre-diabetes,” he noted.
The potential of this innovative technology could revolutionize diabetes screening, making it more accessible and efficient, and ultimately reducing the number of individuals living with undiagnosed type 2 diabetes worldwide. As research progresses, this AI model may pave the way for a future where voice analysis is a standard tool in preventive healthcare.
