When we think of Artificial Intelligence (AI), images of robots from sci-fi movies come flashing, of robots as sidekicks or superheroes, helping their team to make complex decisions and themselves crunching facts at superspeed.
Cut to real life – AI is talked of as a potential game-changer in healthcare. AI may be the answer, from next-generation radiology tools that will do away with the need for tissue samples, to scanning for and diagnosing illnesses like tuberculosis.
A seven-year-long study (2012 – 2019) comparing AI diagnoses with diagnoses made by healthcare professionals showed that AI was able to detect diseases correctly in 87 per cent of the cases, compared to 86 per cent by its human counterparts.
Different & nuanced
However, when it comes to mental health, the answer remains unclear.
Mental health, of course, is a different ball game, with detection and diagnosis being very nuanced. Mental health disorders are not as straightforward as collecting multiple photos of a disorder or illness, like say in tuberculosis, and running them through an AI algorithm to crunch out probabilities based on multiple visual parameters.
“Human psychology or mental health is not as tangible as normal healthcare. For the same set of symptoms in a person, two different psychiatrists may identify two different disorders,” said Nimrod Mon Brokman, Co-Founder of Behavioural Foresight.
Behavioural Foresight, a Bengaluru-based performance and health intelligence company that develops holistic practices, products and enduring hacks for everyday living, has collaborated with a leading Israeli `well-tech’ (or wellness technology) company to help people regain their mental well-being: it works with them on their breath, thoughts, and behaviour. It conducts research into AI-based solutions.
Data & languages
“The most challenging question to solve is which data did you collect and how did you decide to correlate that particular data to certain areas of mental health,” explains Brokman. AI can only be as good as the data it is trained on, and data collected for the purpose of mental health can be challenging because clinical definitions of many mental health related terms are not standardised.
However, there has been significant progress in developing AI algorithms over the past few years. For example, ieso, a mental health clinic headquartered in Milton, Cambridge, the UK, used Natural Language Processing (NLP), an AI method to assess transcripts of therapy sessions. Although the team is still collecting data, it has been able to use data from more than 5.45 lakh therapy sessions to help its clinicians make better decisions.
The story in India is unique. “AI-based mental health apps typically are available as text-based chat bots on the smartphone and rely on NLP techniques to parse user interactions,” says Kashyap Kompella, CEO of AI industry analyst firm RPA2AI Research.
“NLP works well for English and a few other languages but doesn’t work well for most of the world’s languages, including Indian languages. The story is the same with speech recognition techniques as well,” he points out.
Encouraging examples
A 2019 paper reviewed 28 studies on AI and mental health that utilised mood rating scales, brain imaging data, novel monitoring systems and even social media platforms. The studies tried to classify, sub-group and predict mental health issues ranging from depression to schizophrenia.
“Collectively, these studies revealed high accuracies and provided excellent examples of AI’s potential in mental healthcare, but most should be considered early proof-of-concept works demonstrating the potential of using machine learning (ML) algorithms to address mental health questions,” the paper noted.
Gauging user activities
AI works on the method of creating digital phenotypes, which is a model that can infer a user’s behaviour from data collected through Internet of Things (IoT) devices like smartphones, smart watches, and laptop sensors.
“For example, sleep cycles, speech patterns, social interactions, cognitive functioning, physical movements, and several other facets can all be inferred by analysis of the data from smartphones, wearables, and user activity,” explains Kompella.
These digital phenotypes can be especially useful given the vast amount of time that users spend on mobile devices. “It can be beneficial to determine baseline behaviour at the individual level and monitor deviations in behaviour,” he points out.
Therapy from bots
Another wide use case is AI-based bots which can provide automated therapy for users. However, among all of the AI-based tools, there are a few important questions to ask before choosing a tool to diagnose. “The first question is what type of data is being used to drive conclusions. Is it personal data or generic data? Second, how sophisticated are the technologies used and can they help in answering individual needs,” says Brokman.
Researchers believe that as AI techniques evolve, many mental illnesses will be re-defined more objectively than they are currently done through DSM-5 (Diagnostic and Statistical Manual of Mental Disorders). It should be possible to identify illnesses at an initial stage and provide personalised treatments based on the individual.
“On a scale of 1 to 10, I would say AI is at 5. It can currently compensate for a human’s limitation/errors. But it lacks the sensitivity of human contact. Knowing a person can give better answers than statistics,” says Brokman.
