New AI tool: Diagnose type 2 diabetes with 10 seconds of voice | Deutsche Welle | LINE TODAY

(Deutsche Welle Chinese website) Medical diagnostic technology by analyzing speech is becoming more and more precise. Especially in the diagnosis of Parkinson’s disease or Alzheimer’s disease, the analysis of patients’ speech samples can provide valuable basis.

Mental disorders including mental disorders, depression, post-traumatic stress disorder or heart disease can also be diagnosed through speech analysis. Artificial intelligence (AI) can also hear signs of narrowed blood vessels or fatigue. This allows doctors to treat earlier and reduce risk to patients.

Short speech sequences can now tell whether someone has type 2 diabetes with astonishing accuracy, according to a study published in the professional journal Mayo Clinic Proceedings: Digital Health.

unknown risks

The technology is designed to help identify people with undiagnosed diabetes. Approximately 240 million adults worldwide have diabetes without knowing it. According to the International Diabetes Federation, nearly 90% of diabetes cases are type 2 diabetes. People with type 2 diabetes are at increased risk of heart and blood vessel diseases, such as myocardial infarction, stroke, and circulation problems in the legs and feet.

Using speech analysis for diagnosis will make detection easier, as typically people would have to see a doctor for the most common diagnostic tests, which include the fasting blood glucose test (FBG), oral glucose tolerance test (OGTT), and glycated hemoglobin A1C test, which measures the average blood sugar level over a two to three month period.

How does Speech Diagnostics work?

During speech frequency analysis, artificial intelligence analyzes changes in the human voice that are inaudible to the human ear. Simply having a recording of a telephone conversation is sufficient for analysis.

Artificial intelligence can analyze elements such as prosody, rhythm, pauses and pitch of speech. In some clinical cases, patients have unique speech characteristics, such as the pronunciation of the vowel A lasting five seconds.

The human voice can have up to 200,000 different characteristics. Artificial intelligence and corresponding algorithms can find unique articulation patterns in speech sound clips that match certain clinical cases.

Diagnostic accuracy is astounding

The newly developed artificial intelligence tool can identify changes in pitch and volume from 6 to 10 seconds of voice recordings to make a diagnosis. Combined with basic health data such as age, gender, height, and weight, AI can detect whether the person has type 2 diabetes.

And it’s surprisingly accurate: Because male and female voices change differently, so does the accuracy. Despite this, the technology was 89% accurate in diagnosing women and 86% of men.

Distinguishing acoustic features

To program artificial intelligence, a research team led by Ontario Institute of Technology scholar Jaycee Kaufman recorded the voices of 267 people, some of whom did not have diabetes and some who had been diagnosed with type 2 diabetes. For two weeks, the subjects spoke into their smartphones six times a day, speaking just one sentence each time.

From more than 18,000 speech samples obtained in this way, the researchers’ analysis found 14 acoustic features that were significantly different between people without diabetes and people with type 2 diabetes. “Current diabetes detection methods can require significant time, transportation costs and doctor visits. Voice technology has the potential to completely eliminate these barriers,” said Klick Labs researcher Jaycee Kaufman, lead author of the study.

In the future, the Klick Labs team hopes to continue research to see if other conditions, such as high blood pressure and prediabetes, can also be diagnosed using voice detection.

Risks of Speech Frequency Analysis

Proponents of speech analysis as a diagnostic tool often emphasize the speed and efficiency of using speech to diagnose disease.

Yet even if AI-powered tools have provided very precise information, a few speech samples may not be enough to truly make an educated diagnosis. The risk of false-positive results and overdiagnosis also remains high. In the end, a professional doctor still needs to make a diagnosis in person.

provide cluesrather than a medical diagnosis

This principle obviously applies to mental illness as well. The tone of the voice may indicate that he is suffering from depression, but only a thorough examination of the man can really tell.

While AI can use speech analysis to detect, for example, whether a person is speaking in a more garbled or hurried manner than usual, whether this is actually related to attention deficit hyperactivity disorder (ADHD) can only be determined by a medical professional. Make a diagnosis.

Risk of abuse cannot be ruled out

Critics and data protection advocates have also repeatedly pointed out that there is a huge risk of abuse when using speech analytics, such as by employers or health insurance call centers. There is a risk that speech analysis systems may be used without explicit consent, and that customers or employees may be disadvantaged by personal health information.

Likewise, sensitive medical information may be more susceptible to being leaked, hacked, sold, or otherwise misused. Ultimately, it is not the scientific community but the political community that can establish clear rules and limits for speech analysis as a diagnostic tool.

© 2023 Deutsche Welle Copyright Statement: All contents of this article are protected by copyright law and may not be used without special authorization from Deutsche Welle. Any misconduct will result in recovery of damages and criminal prosecution.