NEW YORK, BRONX (ORDO News) — Canadian medical researchers have made significant strides in the battle against Type 2 diabetes by training a machine-learning AI to accurately predict the condition from just six to 10 seconds of a patient’s spoken voice.
The AI model identified 14 acoustic features that distinguish between non-diabetic and Type 2 diabetic individuals. It analyzed vocal characteristics, including subtle changes in pitch and vocal intensity that are often imperceptible to human ears.
When combined with basic health information, such as the patient’s age, sex, height, and weight, the AI proved to be highly effective in diagnosing Type 2 diabetes.
Researchers found that sex played a crucial role in diagnosis. The AI demonstrated an 89% accuracy rate for diagnosing Type 2 diabetes in women and a slightly lower but still impressive accuracy rate of 86% for men.
The remote, automated diagnosis offered by this AI model has the potential to significantly reduce the cost and improve the accessibility of Type 2 diabetes diagnosis, as it eliminates the need for in-person testing.
Traditional diagnostic tests, such as the glycated hemoglobin (A1C) test, fasting blood glucose (FBG) test, and oral glucose tolerance tests (OGTT), require patients to visit healthcare providers, which can be costly and time-consuming.
The implications of this AI are profound, especially in the context of the global diabetes epidemic.
According to the International Diabetes Federation, nearly half of adults living with diabetes, around 240 million individuals, are unaware that they have the condition.
This AI tool could be a game-changer, significantly improving early diagnosis and treatment for those at risk.
Researchers from Klick Labs collaborated with faculty at Ontario Tech University in Canada to develop this AI model.
They trained the AI using voice recordings from 267 test subjects, including both non-diabetic individuals and those previously diagnosed with Type 2 diabetes.
The participants recorded a specific phrase on their cell phones multiple times daily for two weeks, resulting in a dataset of 18,000 recordings.
The AI focused on 14 acoustic features that consistently distinguished between the two groups, with four features proving particularly useful for prediction.
These findings have the potential to revolutionize healthcare practices by providing a non-intrusive and affordable digital screening tool for Type 2 diabetes.
The researchers aim to replicate their study’s findings and expand their voice-diagnosing research to other medical areas in the future, such as prediabetes, women’s health, and hypertension.
This innovation demonstrates the potential for AI to significantly improve early disease detection and provide more accessible healthcare solutions.
—
Online:
News agencies contributed to this report, edited and published by ORDO News editors.
Contact us: [email protected]
Our Standards, Terms of Use: Standard Terms And Conditions.
To eliminate any confusion arising from different time zones and daylight saving changes, all times displayed on our platforms are in Coordinated Universal Time (UTC).