Diabetes is a chronic condition that affects millions of people worldwide. It occurs when the body cannot produce enough insulin or use it effectively to regulate blood sugar levels. Diabetes can cause serious complications such as heart disease, kidney failure, nerve damage, and vision loss. Therefore, early diagnosis and treatment are crucial to prevent or delay these outcomes.
However, many people with diabetes are unaware of their condition, especially in low- and middle-income countries where access to health care and testing facilities may be limited. According to the International Diabetes Federation, about 240 million adults have diabetes and do not know it. This poses a major challenge for public health and individual well-being.
Fortunately, advances in artificial intelligence (AI) and voice analysis may offer a simple and convenient way to screen for diabetes. A recent study published in Mayo Clinic Proceedings: Digital Health showed that AI can detect diabetes with a 10-second voice sample with high accuracy. The study was conducted by researchers from Klick Applied Sciences, a digital health innovation lab.
How does voice analysis work?
Voice analysis is a technique that uses AI and machine learning to examine various features of human speech, such as pitch, intensity, melody, cadence, and pauses. These features can reveal subtle changes in the voice that are not perceptible to the human ear, but may indicate certain health conditions or symptoms.
For example, voice analysis can help diagnose diseases such as Parkinson’s, Alzheimer’s, depression, post-traumatic stress disorder, and heart disease. Voice analysis can also detect signs of constricted blood vessels or exhaustion.
The researchers from Klick Applied Sciences used voice analysis to identify vocal patterns that are associated with type 2 diabetes, the most common form of diabetes. Type 2 diabetes occurs when the body becomes resistant to insulin or does not produce enough of it. This causes high blood sugar levels, which can damage various organs and tissues.
How does AI detect diabetes with a voice sample?
The researchers recruited 267 participants from India, of whom 75 had type 2 diabetes and 192 did not. The participants used a smartphone app to record a single fixed phrase for six to 10 seconds up to six times a day for two weeks. The app collected a total of 18,465 voice recordings, which were then analyzed by AI algorithms.
The AI algorithms measured 14 different acoustic characteristics of the voice recordings, such as pitch, intensity, and perturbation. They also considered basic health data of the participants, such as age, gender, height, and weight. The AI algorithms then used an ensemble model to classify the voice recordings as either diabetic or non-diabetic.
The results showed that the AI model was able to detect diabetes with 89% accuracy for women and 86% accuracy for men. The model also identified the most important vocal features for diagnosing diabetes. For women, these were related to pitch, such as the average pitch, the deviation from the average pitch, and the relative average fluctuations in pitch. For men, these were related to intensity, such as the average intensity, the deviation from the average intensity, and the relative average perturbations in pitch.
What are the benefits and limitations of voice analysis for diabetes screening?
The researchers believe that voice analysis could provide a simple, non-invasive, and cost-effective way to screen for diabetes, especially in low-resource settings where access to blood tests may be limited. Voice analysis could also reduce the stigma and inconvenience associated with blood tests, and encourage more people to seek diagnosis and treatment.
However, voice analysis also has some limitations and challenges. For instance, the accuracy of the AI model may vary depending on the quality of the voice recordings, the background noise, the language, the dialect, and the cultural context. The AI model may also need to be validated and calibrated for different populations and settings. Moreover, voice analysis may raise ethical and privacy concerns, such as how the voice data will be collected, stored, shared, and used.
Therefore, voice analysis should not be seen as a replacement for blood tests, but rather as a complementary tool that can help identify people who may have diabetes and need further confirmation and care. Voice analysis could also be integrated with other digital health solutions, such as telemedicine, mobile apps, and wearable devices, to provide more comprehensive and personalized care for people with diabetes.