A new AI-driven tool developed by the Luxembourg Institute of Health aims to make diabetes screening faster and more accessible by analysing vocal patterns associated with the condition.

A project by the Luxembourg Institute of Health (LIH) highlights the critical role of research in medicine and the potential of AI to revolutionise healthcare. The initiative focuses on detecting type 2 diabetes through voice analysis, offering a non-invasive and innovative approach to identifying the condition.

Led by Dr Guy Fagherazzi, a team of LIH researchers has developed AI-powered technology capable of analysing subtle vocal changes in individuals with diabetes – changes that are often imperceptible to the human ear. In an interview with our colleagues from RTL Télé, Dr Fagherazzi explained: "People with diabetes have a different voice compared to those without the condition." According to the researcher, this can be attributed to various factors, such as hyperglycaemia, fatigue, and neuropathies, all leading to "a generally hoarser voice".

The research leverages data from the international Colive Voice study, which has collected over 8,000 voice samples in multiple languages. For the specific analysis of type 2 diabetes, the team used a subset of English-language recordings from approximately 600 individuals in the United States. Participants were asked to read a 20-30 second text, and their recordings were then analysed by AI to identify patterns associated with diabetes.

Dr Fagherazzi noted that the AI system successfully distinguished between individuals with and without type 2 diabetes by assigning distinct scores to each group.

With an estimated 800 million people worldwide living with diabetes – half of whom remain undiagnosed – this technology could provide a quick and accessible tool for identifying individuals at risk.

However, Dr Fagherazzi emphasised that voice analysis is not intended to replace traditional diagnostic methods, such as blood tests and glucose level assessments. Instead, it aims to serve as a preliminary screening tool to flag high-risk individuals.

"Currently, our technology achieves an accuracy rate of around 75%", Dr Fagherazzi stated, adding: "Hopefully, over time we'll get even better."

To enhance the tool's precision, the research team is actively seeking additional voice samples and encourages volunteers to participate in the study via the www.colivevoice.org website.

Full report by RTL Télé (in Luxembourgish and French)