Dialing in on Diabetes Diagnoses

Type 2 diabetes is a chronic metabolic disorder characterized by high blood sugar levels, resulting from the body's ineffective use of insulin or the inability to produce enough insulin. This condition often develops slowly over time and can go unnoticed for an extended period. It is commonly associated with lifestyle factors such as obesity, sedentary behavior, and poor diet, although genetic and environmental factors can also contribute.

The effects of Type 2 diabetes on patients can be far-reaching and severe. Over time, elevated blood sugar levels can lead to various complications, including damage to the eyes, nerves, kidneys, and blood vessels. Patients with uncontrolled diabetes are at a higher risk of developing cardiovascular diseases, such as heart attacks and strokes. Additionally, the disease can lead to complications like diabetic neuropathy, which causes numbness, tingling, or pain in the extremities, as well as foot ulcers that may result in amputations if left untreated.

There are many ways to diagnose Type 2 diabetes. The A1C test, which measures the average blood sugar levels over the past two to three months, is a commonly used diagnostic tool. A fasting blood glucose test measures blood sugar after an eight-hour fast, providing an indication of how the body processes sugar. Another diagnostic method is the oral glucose tolerance test, which involves drinking a glucose-rich solution and measuring blood sugar levels before and two hours after consumption. While these tests are essential for early detection and management, the reliance on lengthy processes, laboratory settings, and medical professionals often poses a barrier to timely testing for many individuals.

As a result, many people with undiagnosed Type 2 diabetes may experience prolonged periods without treatment, leading to the progression of the disease and potentially severe health consequences. But a surprising new diagnostic tool has recently been announced by researchers at Klick Applied Sciences that may change this present situation. Rather than measuring blood sugar levels, the team made the seemingly unlikely choice to analyze short voice samples. By speaking a few sentences into a smartphone, their system can diagnose Type 2 diabetes with 89% accuracy in women, and 86% accuracy in men.

Because it is noninvasive, convenient, and inexpensive, voice-based diagnoses have garnered a lot of attention in recent years. In theory, anything that impacts the respiratory system, nervous system, or larynx could be detected by analyzing the characteristics of one’s voice. Since it is suspected that diabetes can change the properties of one’s vocal chords, and also can cause neuropathy and myopathy, the team believed that a unique signature associated with diabetes may be hidden in certain vocal characteristics.

To test out this theory, the researchers collected more than 18,000 voice recordings from hundreds of diabetics and non-diabetics. After collecting these recordings over a period of two weeks at different times of day, fourteen different parameters, like pitch and intensity, were extracted. These parameters were used to train multiple machine learning models to classify voice samples as diabetic or non-diabetic.

The best performing model proved to be a Naïve Bayes classifier, which correctly identified voice samples with nearly 90% accuracy. The subtle vocal traits that it identified were generally inaudible to the human ear, so would not be detectable without a system such as this.

By removing barriers like high cost and inconvenience, this breakthrough could potentially revolutionize the management of Type 2 diabetes in the future. By diagnosing those with the condition shortly after it surfaces, patients can be directed to make lifestyle changes which can significantly improve symptoms and prevent serious complications before they have a chance to occur. Further down the road, the team hopes to tackle the diagnosis of more medical conditions, like high blood pressure, with a similar tool.