Study Finds AI Can Diagnose Type 2 Diabetes Just Listening to Your Voice

A recent study from the Mayo Clinic has found that voice analysis may be a promising tool for diagnosing type 2 diabetes mellitus (T2DM).

Recent research from the Mayo Clinic has delved into the potential of voice analysis as a tool for diagnosing type 2 diabetes mellitus (T2DM). The study, titled “Acoustic Analysis and Prediction of Type 2 Diabetes Mellitus Using Smartphone-Recorded Voice Segments,” aimed to discern the differences in voice recordings between individuals with T2DM and those without the condition.

Methodology and Findings

The study involved 267 participants from India, diagnosed either as nondiabetic or with T2DM based on American Diabetes Association guidelines. Over a span of two weeks, participants used a smartphone application to record a fixed phrase multiple times daily, resulting in a total of 18,465 recordings. From these recordings, 14 distinct acoustic features were extracted to analyze the differences between the two groups and to develop a prediction methodology for T2DM status.

The results revealed significant differences in the voice recordings of both men and women with T2DM compared to those without the condition. For women, the most predictive features were pitch, pitch SD, and relative average perturbation jitter. For men, intensity and 11-point amplitude perturbation quotient shimmer were the most indicative. When these features were combined with age and BMI data, the prediction models achieved accuracies of 0.75 for women and 0.70 for men.

An accuracy of 0.75 (or 75%) means that for women, the model correctly predicted the presence or absence of T2DM in 75 out of 100 cases. Similarly, an accuracy of 0.70 (or 70%) for men means that the model was correct in 70 out of 100 cases.

Implications and Potential Applications

The findings suggest that vocal changes are evident in individuals with T2DM compared to those without the condition. Voice analysis, especially when combined with other risk factors, shows promise as a prescreening or monitoring tool for T2DM. This non-invasive, cost-effective method could be particularly beneficial in remote or underserved communities with limited access to healthcare services.

Voice synthesis, which is a complex interplay of the respiratory system, nervous system, and the larynx, can be influenced by various factors. In T2DM, sustained periods of high blood glucose can affect the elastic properties of the vocal cords. Long-term elevated glucose levels can lead to peripheral neuropathy and myopathy, both of which can influence voice quality. Additionally, T2DM has been associated with psychological disorders like depression and anxiety, which can also result in vocal changes.

The overarching goal of this research was to assess the feasibility of using voice as a predictive tool for T2DM. Preliminary results have been promising, but more extensive data is needed, especially from age-matched and BMI-matched populations. The study also emphasized the importance of simulating real-world scenarios by collecting data through mobile applications and analyzing voice features separately for men and women.