Salud Diabetes

Diabetes News for the Spanish Speaking Community

AI Voice Analysis as a Novel Tool for Diabetes Screening

Published · Updated

Original Article: https://www.medscape.com/viewarticle/ai-voice-analysis-diabetes-screening-shows-promise-2024a1000ggw

Promising Results from AI-Driven Voice Algorithm

At a recent research meeting in Europe, researchers introduced a voice analysis AI model that aligns closely with the American Diabetes Association’s (ADA) risk test for detecting type 2 diabetes (T2D). This AI algorithm showed “excellent agreement” with the ADA’s risk test, achieving a 66% accuracy rate in women and 71% in men. Notably, it demonstrated a 93% correlation with ADA’s questionnaire-based screening method, signaling a potential alternative to traditional methods.

Study Overview and Findings

In the present study (Colive Voice project), the sample was limited to English-speaking adults in the U.S., both with and without diabetes. The team developed and cross-validated algorithms to analyze voice recordings of roughly 25 seconds, considering health data such as age, sex, BMI, and blood pressure. The voice samples—323 from women (162 with T2D and 161 without) and 284 from men (142 with T2D and 142 without)—underwent AI analysis for predictive capacity. The model analyzed vocal features like pitch and intensity, capturing up to 6000 vocal characteristics, with a deep learning approach focused on a subset of 1024 key features. The algorithm appears to be a promising first step in non-invasive diabetes screening.

Challenges and Future Prospects

The study’s researchers emphasize the need for further testing across more diverse populations. They also point out the importance of expanding the algorithm to detect prediabetes, which could offer earlier intervention opportunities. Another avenue of research involves exploring the relationship between glucose levels and voice characteristics, as other studies have hinted that changes in vocal fold tension, possibly influenced by glucose, could be significant. 

In conclusion, while the AI voice analysis tool has shown significant promise, more research is needed to assess its reliability across different populations and its potential to predict glucose levels and detect prediabetes.