“You are sick!” Between words, AI can diagnose mental disorders.

In “To Youth”, Chen Xiaozheng blurted out because he heard Zheng Wei telling him that “you are mentally sick!” However, you don’t laugh. People convey their thoughts, tone of speech, choice of words, and length of sentences through the content and manner of speaking. They are all key clues to understanding each other’s thoughts.

Via serious gossip

In the field of mental illness, when doctors or psychologists examine patients, they will listen to these language signals to obtain the patient's health status, and then rely on the doctor's past experience to guide the patient's judgment.

Now, researchers are now using the same machine learning function to diagnose patients with mental disorders.

In 2015, a research team developed an artificial intelligence model that successfully analyzed the text of a group of young people's conversations and predicted who among them would have mental confusion. (Insanity is a major feature of schizophrenia). This model focuses the problem on speech twitches that can manifest insanity, such as the use of short sentences, expression confusion, frequent use of words such as "this", "one", and confusion between sentences.

In Hitchcock’s classic movie “The Psycho”, a son with mental illness mumbled before he was arrested.

Now, Jim Schwoebel, CEO and engineer at NeuroLex Diagnostics, wants to create a tool based on the above research to help primary care physicians check whether their patients have mental disorders. The company's products can record patient consultations through smart phones or other devices (walls mounted outside the line of sight).

Then, the product can use the artificial intelligence model to find the language clues from the text of the patient's speech and display its findings in digital form. Just like blood pressure readings, the psychiatrist can use this reading as a basis for diagnosis. And, because the algorithm is continuously trained and learnt in more and more patients, the readings it gives can better reflect the patient's mental state.

In addition to screening for schizophrenia, Schwoebel was also awarded by the American Psychiatric Association for another idea. NeuroLex hopes to develop a tool to help mentally ill patients who have been admitted to hospital. Instead of helping doctors diagnose symptoms of mental confusion from a single case, this artificial intelligence can track their course of treatment through long-term testing of patient conversations.

For Schwoebel, this work also has a private purpose. He thinks this method may solve the problem of schizophrenia treatment faced by his brother. Before his brother’s first episode of mental illness, he showed a number of worrying anomalies, such as short or one-word responses, tending to say “here,” “then,” and so on.

Schwoebel said: "After my brother's first mental illness experience, the previous performance makes sense."

Before Schwoebel's elder brother turned to the psychiatrist and eventually got a diagnosis, he had at least 10 primary care consultations. After this, his brother was caught in the failure of drug treatment again and again. Over the years, his brother has experienced three episodes of mental illness and finally got the correct diagnosis and effective health treatment. Repeated failures in drug treatment have left Schwoebel wondering how to get patients to get the correct prescriptions as soon as possible.

In order to find out, NeuroLex plans to conduct a “front-to-back study” to study how the language patterns of patients with mental disorders who have been admitted to hospital change during the course of their illness. Ideally, artificial intelligence will analyze the samples recorded by patients during mental rehabilitation and compare which drugs are more effective, thereby reducing the time spent in hospital treatment.

If the patient's conversation shows less signs of depression or bipolar disorder after using a medication, this tool can help explain that the medication is working. If the patient's conversation does not show a significant change, artificial intelligence will suggest that other drugs be tried immediately to reduce the patient's pain.

And, once the artificial intelligence collects enough data, it can recommend suitable medicine according to the case of other patients with the same speaking pattern. For decades. The automatic diagnosis method has been applied more widely in the field of medicine. One company claims that their algorithms are 50% more accurate than human radiologists in recognizing lung cancer.

This possibility of using a more objective and quantitative assessment to help psychiatrist clinicians diagnose the disease appealed to Arshaya Vahabzadeh, a psychiatrist who lived at the Saskatchewan General Hospital, who is now a senior advisor to the start-up accelerator fund co-founded by Schwoebel. He said: "Schizophrenia involves a series of observable or elicitable symptoms, rather than an all-encompassing diagnosis. If there is a large enough data set, artificial intelligence can be based on the observed common characteristics of patients, Divide some schizophrenia-like diagnoses into more precise and more helpful categories. I think these data can help our team reclassify some of the conditions that we could not have done before."

Vahabzadeh added: "As with other drug interventions, the help of artificial intelligence needs to continue to be researched and confirmed. This is what I want to highlight. Schwoebel thinks that too." Although research on prediction of mental disorders shows that language analysis can be reasonably correct Predicting mental confusion, but this is still only a study. No one has published proof of opinion on depression or bipolar disorder.

Machine learning is a hot area, but it still has a long way to go. There are many aspects of the machine itself and the machine that need to be studied. For example, Siri has been addressing Scottish users' directives and problems for years. For the treatment of mental illness, a small problem like this may cause a disaster. "If you tell me that a technology has a 20% error rate in practice or only 80% accuracy, I will not apply it to patients," Vahabzadeh said. (Editor's Note: Prerequisites for the application of artificial intelligence products generally require more than 95% accuracy, and in practice, the span from 95% to 99.5% is often considered to be an important watershed, and after reaching this accuracy, the artificial intelligence The addition of human-assisted judgment will greatly increase work efficiency.)

When considering the patient's age, gender, race, ethnicity or region, this risk will be even more unbearable. A sample of the language in which an artificial intelligence is trained to analyze is all from the same group of people. Normal samples outside this group may be judged to be abnormal.

"If you come from a particular cultural group, you may talk softly and have lower pitches, then AI may mistakenly think you have depression," Schwoebel said.

But Vahabzadeh still believes that such technology will one day help clinicians treat more people and treat patients more effectively. More importantly, in view of the insufficiency of mental health rehabilitation personnel in the United States, if people do not take cost-effective solutions, we have to use technology in some way to support the doctors.

Via The Atlantic

Recommended reading:

Surveys show that sex robots can reduce the spread of sexually transmitted diseases and illegal transactions. Do you want to try it? | New wisdom

Take skin shots with SkinVision to detect skin conditions