Statistics show that there are 1 billion people suffering from mental problems worldwide. Depression is a serious illness that undermines people’s health.
Chinese artificial intelligence startup Voice Health Tech has started a joint project with a group of researchers led by Yue Weihua, a professor at Peking University Sixth Hospital, on the use of speech for depression screening and evaluation.
The project has now entered the clinical research stage. In November 2022, the team published a clinical research paper in Frontiers in Psychiatry, an international medical journal. The team uses a method based solely on deep learning for acoustic signal processing, making it possible to diagnose depression from a roughly 30-second speech via smartphone.
The method’s sensitivity and specificity – measures used to determine the accuracy of a test – are said to have reached 82.14% and 80.65%, respectively. Sensitivity is the ability of a test to identify patients with a disease, while specificity is the ability to identify people without the disease.
Despite using stricter evaluation criteria called DSM-5, or Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, the performance shown by this deep learning model in clinical research is said to have exceeded that of related Western companies.
The project’s speech model was prepared using a data set comprising up to 43,000 clinical conversations. It was also tested and studied using an independent validation data set.
Conversations collected from patients, who were all diagnosed based on DSM-5, was recorded with different smartphones in various settings. He Gongcheng, Voice Health Tech co-founder and chief medical officer, said the quality of the data set might be the highest in the field at this stage.
Statistics show that there are 1 billion people suffering from mental problems worldwide. Depression is a serious illness that undermines people’s health.
According to a World Health Organization report, more than 300 million people are estimated to be suffering from depression, with an average incidence rate said to be 4.4%. In China, the lifetime prevalence of depression is said to be as high as 6.8%. Depression is also projected to become a major disease burden worldwide by 2030.
As things stand now, many patients with depression are wrongly diagnosed or cannot receive necessary support due to factors such as medical accessibility. Depression has so far been diagnosed by psychiatric experts, who interview and evaluate patients based on certain criteria to determine who suffers from depression.
But a shortage of psychiatrists is making diagnosis more difficult in China. As of 2017, China had 27,000 psychiatrists – or two psychiatrists per 100,000 people.
According to WHO data, Russia and the US have 11 and 12 psychiatrists, respectively, per 100,000 people. Although the number of psychiatrists in China increased to 40,000 in 2020, their presence remains comparatively small, considering the country’s population.
Furthermore, some patients with depression are not aware of the seriousness of their symptoms or try to hide them from doctors. That is why diagnostic errors are often made under traditional diagnostic methods.
In recent years, AI technology for mobile devices has been playing a significant role in monitoring physiological and psychological data in a non-invasive and continuous manner.
At the same time, technologies such as semantic recognition and machine translation have matured gradually, while acoustics and voice command processing has made progress. As a result, machine learning has emerged as a new field in diagnostics. Certain acoustic features – including glottal, spectral and prosodic – are used to identify depression.
Machine learning makes it possible to detect objective changes in acoustic features that humans cannot perceive with their ears. It also shows potential in terms of detecting mental disorders, including depression.
Originally published at Free Malaysia Today