Beyond subjectivity: How technology aims to improve psychiatric diagnoses

A study published in Nature in 2024 identified six depression biotypes based on the brain dynamics of more than 800 patients. Functional magnetic resonance imaging (fMRI) showed, for example, that some reported cognitive overactivity, while others showed a pattern of poor performance in the neuronal circuitry that controls attention. In their therapeutic correlates, the former responded better to certain types of antidepressants, and among the latter, psychotherapy did not seem to work too well. In another group of patients, hyperconnectivity of brain circuits and high benefit from behavioral interventions were the norm. The other three biotypes also presented their own neurological characteristics and different levels of therapeutic response.

Martien Kas, a Dutch researcher and president of the European College of Neuropsychopharmacology, cites this study to illustrate how much technology can help refine psychiatric diagnoses. This is a field of medicine in which — unlike other specialties — patient diagnosis is still carried out almost exclusively on the basis of highly subjective information: what the patient says is happening to them (their symptoms) and what the doctor interprets based on this account.

Kas sums up his optimism: “There is an excellent opportunity to take advantage of emerging knowledge about the brain to use this quantitative information to make better diagnoses.” He alludes to a pioneering analysis — published in 2016 in the American Journal of Psychiatry — that, using a wide range of brain markers, classified three biotypes for people suffering from psychotic disorders such as schizophrenia.

The authors acknowledged their aim to go beyond — guided by objectivity — the heavy reliance on the DSM manual (the bible of psychiatric diagnosis) when it comes to labeling the suffering of people with mental health issues. The DSM, in its fifth edition (updated in 2022), lays out an overwhelming catalog of nearly 300 disorders with symptoms that, according to some critics, are vague and tend to overlap.

Better differentiating pathologies that sometimes co-occur in similar emotions and thoughts is one of the main goals of Kamilla Miskowiak, a researcher at the University of Copenhagen in Denmark, who is applying virtual reality tools to obtain more accurate assessments. According to a review published in The BMJ in 2024, the false positive rate for depression could be higher than 60%, bipolar disorders are often confused with psychosis, and schizophrenia diagnoses change, over the medium term, for a range of 30-50% of patients.

In a recent pilot study published in European Neuropsychopharmacology, Miskowiak and colleagues measured skin conductance in response to immersive virtual reality scenarios (a crowded elevator, a crying baby) in 100 patients with schizophrenia, bipolar disorder, and borderline personality disorder (BPD). Differences in this biomarker — considered by Miskowiak highly effective for assessing “individual agitation” — were notable among the three groups.

According to Miskowiak, “virtual reality is a very promising technology for more accurate diagnosis.” This is due to its low cost and because, in disorders like BPD, “the person does not usually cooperate with the doctor due to their difficulties with social interaction.” In any case, Miskowiak admits that we are still far from being able to use it reliably in psychiatrists’ offices.

An app that ‘sees’ post-traumatic stress

Some companies are already marketing technological products that are supposedly useful for narrowing down the sometimes vague nature of each disorder. Texas-based Senseye advertises on its website what it calls “the world’s first diagnostic platform for objectively assessing mental health.” Its methodology focuses on an individual’s eye activity (gaze direction, pupil dilation, iris biomechanics) during specific visual tasks. For now, its cell phone application is only used to diagnose post-traumatic stress disorder (PTSD). According to the company, its product is at least as effective as the CAPS-5, the most widely used scale for determining whether someone has PTSD.

Jessica Jackson, chair of the American Psychological Association’s advisory committee on technology and mental health, is cautious, stating categorically that “we are not yet capable of making better diagnoses thanks to technology.” Jackson mentions promising tools that attempt to detect depressive traits in the voice using large language models — the foundation of generative artificial intelligence (AI) — although for now she considers these and similar initiatives to fall into the realm of “guesswork.”

Martien Kas admits that we still “do not fully understand the underlying mechanisms in psychiatry” and that we do not fully know “what, how, and when to measure,” although progress has accelerated in recent times. Jackson adds that mental disorders often involve a series of “environmental and biological” causes that are difficult to decipher, with relational complexity adding another hurdle for accurate diagnosis. Furthermore, she points out, there are cultural factors whose ignorance can mislead even the best psychiatrist: “For example, anger as a symptom of depression among the Black community in the U.S.”

The arrival of generative AI

For Kas, it is precisely in this murky mix of factors, in the “bilateral interaction between the brain and the environment, the understanding of which requires collecting a large amount of multimodal data,” that AI can shed the most light. A review published in Nature in 2022 found highly variable accuracy (21% to 100%) in AI models that integrate all kinds of information (neuroimaging results, psychological evaluations, etc.) to decide whether someone has disorders such as obsessive-compulsive disorder or bipolar disorder.

Another review published last February in Cambridge University Press observed that AI-based mental health diagnoses often rely on random trees, a learning algorithm that uses a weighted analysis of a data amalgamation to better understand the forest (the specific pathology). The study, in which Spanish researcher Pablo Cruz participated, also concluded that AI has the potential to improve psychiatric practice both predictively (by alerting people to risk factors that lead to early intervention) and by more accurately suggesting therapeutic alternatives.

Cruz remains cautious: “We are [in applying AI in psychiatry] at an early stage, although the situation could change enormously in 10 or 15 years.” He adds that AI — or technology as a whole — does not necessarily need to fully replace the clinical eye of doctors, but can simply “help increase the likelihood of correct diagnosis” and ensure it is not “a one-shot game.”

Jessica Jackson also predicts that, in the future, technological tools will carry “at least as much weight” as a doctor’s subjective judgment, especially in severe illnesses like schizophrenia (where there is greater potential to isolate specific biomarkers) and less so in more common, milder disorders such as anxiety. As Kas notes, anxiety (or its close relative, fear) serves an evolutionary function that complicates distinguishing pathology from life itself— even when using the most sophisticated technology.

Sign up for our weekly newsletter to get more English-language news coverage from EL PAÍS USA Edition