Machine Learning in Mental HealthTech

Mental health illnesses affect millions of people worldwide. Only in the US, according to the 2020 statistic, 21% of adults experienced mental illness, which is 52.9 million people and represents 1 in 5 adults. Anxiety disorders are the most common, affecting 19.1% of the population ― an estimated 48 million people. They are followed by people experiencing major depressive episodes, which make up 8.4% of the US population, or 21 million people. Yet, only 46.2% of U.S. adults with mental illness received treatment in 2020. This is less than half. 

There are multiple reasons for that. Even in 2022, the stigma around mental health issues is still real. Many individuals are embarrassed and uncomfortable talking to a psychiatrist or therapist, and admitting they have a problem. Many individuals don’t even know about their vulnerabilities: while the awareness of mental health has skyrocketed in the past decade, it still hasn’t made up for all the decades we’ve lived without it. Millions of people still can’t pinpoint when something is wrong with their mental health and decide to seek help.

There are also multiple barriers for patients with mental health disorders that have decided to seek help. Accessing and receiving adequate care isn’t a simple task even in developed countries. One has to figure out when and where to seek help, find a provider, and, most importantly, have financial resources, time, and transportation to attend appointments. 

These challenges, exacerbated by the COVID-19 pandemic of 2020, led to the rise of mental health technology. Different healthcare IT solutions, such as mental health platforms, bots, and apps achieved very good results: they were cheaper, easy to use, didn’t require the patients to go anywhere, and allowed higher engagement. In one of the studies with participants suffering from depression, those who used Woebot ― a mental health help bot powered by AI ―  experienced close to a 20% improvement in just two weeks. One reason for Woebot’s success was the high level of participant engagement. Most participants were talking to the bot nearly every day. This level of engagement simply isn’t possible with in-person counseling.


How machine learning is used in Mental HealthTech?

Woebot, a mental health bot that we mentioned earlier, was an example of a bot powered by artificial intelligence. How do artificial intelligence and machine learning fit in with mental health tech?

Essentially, machine learning is a type of AI (artificial intelligence) technology. Machine learning works in a way that a machine is given lots of data and examples of which kind of output it should produce when a particular input is received. After spending some time learning, the machine can get quite good at autonomously performing a task, making connections, and identifying patterns in large datasets. It can then be used in several ways: 


Developing treatment plans

Treatment plans developed by machines that have soaked in huge amounts of data can help clinicians make better decisions. Machines analyze who responds to what treatment, identify key biomarkers, and determine relevant sub-types of different disorders and which treatments are most effective to deal with them. Then, the algorithms may assist in tracking the effectiveness of a treatment plan. 

Basically, machine learning helps personalize treatment, while normally diagnoses are based on group averages and statistics over populations. 


Predicting crises

Machine learning algorithms could help determine key behavioral biomarkers for mental health disorders before the disorders set in. This will help clinicians predict who may be at risk of a particular disorder. Mental health help will then come before the person has even developed a disorder. Prevention, which is so prominent in other healthcare fields, will finally reach mental health. 

Technology that fulfills these two tasks already exists. Kintsugi, a mindful talk therapy software, uses machine learning algorithms to detect depression and anxiety signs by speech biomarkers. Their voice biomarker API platform called KiVA integrates with clinical call centers, telehealth platforms, and remote patient monitoring apps to provide clinicians with real-time scoring on patients’ mental health. This technology, as the company claims, detects voice biomarkers that indicated markers of depression and anxiety based on how patients speak rather than natural language processing, which makes this technology applicable in any country and in any language. The correctness of the scores is backed by the insights from its extensive and constantly evolving global dataset. As long as your machine-learning software has enough data, it can do anything. 


What is the nearest future Mental HealthTech?

Machine learning becomes unstoppable when you combine it with wearable devices and smartphones, so most research around mental health tech is going in this direction. At MIT, Rosalind Picard, who is an MIT scientist, collaborates with clinicians to develop tools for mental health care delivery. They gather data on participants’ skin conductance and temperature, heart rate, activity levels, socialization, personal assessment of depression, sleep patterns, and more to develop machine learning algorithms for mental health care. This algorithm will be able to intake this tremendous amount of data and make it meaningful — identify when exactly an individual may be struggling and propose suitable treatment plans. 

Researchers say that eventually, individuals will have the opportunity to access information that is evidence-based and personalized, and shows their personalized risks and how to deal with them. For example, imagine a wearable device or a smartphone can one day show that the user has recently been sleeping less, staying inside their home more, and has a faster-than-usual heart rate. These changes are usually subtle and often invisible to the individual in question and also to their loved ones. The device, on the other hand, is powered by machine learning. It can make sense of these data, and map the results onto the individual’s past experiences and the experiences of other users. Then, after making conclusions, the device can encourage the individual to engage in certain behaviors that are helpful in this case. These might be the behaviors that help other users or that have improved the individual’s well-being in the past. The device might also suggest reaching out to their physician. 

What are the challenges of using machine learning in Mental HealthTech?

Often, it’s easier to diagnose than it is to treat. This applies to in-person counseling, and it also applies to machine-learning technology. It’s important to not only warn the user that they are heading towards deep depression: this may discourage the user and make them even more hopeless, exacerbating and speeding up the coming of the depressive episode. Instead, the tool must state the reason the user is feeling down which must be backed by data. For example, the data related to their sleep pattern, social activity, and physical activity. Then, there should be a recommendation to change that. 

Data privacy and informed consent are also vital for this kind of technology, as the information is sensitive and could lead to terrible consequences if leaked. 

Finally, it’s important that whichever technology appears on the other end, it’s not seen as a complete substitute for in-person therapy. Often, the years of learning and practice and the social communication that the therapist provides are irreplaceable. Besides, the patient might require drug prescriptions, and AI is not legally allowed to provide that. At least, not yet.

Contact Us
Contact Us