Deep Learning Approach Shows Promise for Automatic Assessment of Schizophrenia Using Wearable Technology

Schizophrenia and bipolar disorder share many traits. Both are severe mental illnesses that significantly impact quality of life. Both are associated with autonomic nervous system dysfunction, which can be assessed through heart activity analysis. (Indeed, there is an ongoing debate whether the two diseases, which are classified as distinct based on patients’ symptoms, actually represent a continuum). Both can be challenging to diagnose. And early detection of both diseases is crucial to better outcomes.

Diagnosis in the prodromal stage — after the appearance of early warning signs and symptoms, but before the full onset of schizophrenia, which leads to a gradual decline in emotional, behavioral and cognitive functioning — is uncommon. Later diagnosis often results in admission to a psychiatric unit, higher costs for patients and an increasing burden on the healthcare system.

Psychiatry is one of the few areas in medicine that does not have the benefit of laboratory blood testing, including genetic profiling, that measures biomarkers for diagnosis and disease monitoring. Most serious mental disorders are diagnosed based on clinical interviews by trained physicians who interpret subjective and variable sets of symptoms reported by patients.

The diagnostic approach described and tested in a recent study would break that mold. The study, which was published Sept. 3, 2025, in PLOS Computational Biology, used heart rate data, which might not meet some strict definitions of biomarker but would still be a marked departure for a psychiatric condition. Based on results from two groups of 30 patients each — one whose members had been diagnosed by psychiatrists with schizophrenia or bipolar disorder (the data was analyzed collectively) and the other control — it classified individuals with each of those diseases with about 80% accuracy.

“Although not a replacement for clinical diagnosis, heart monitoring combined with artificial intelligence could help detect signs of mental illness, leading to faster interventions,” the authors write, adding that their work could ultimately lead to the development of easy-to-use tools that could support doctors and improve patient care.”

Corresponding author Kamil Książek, Ph.D., a postdoctoral researcher in the Group of Machine Learning Research at Jagiellonian University in Kraków, Poland, and colleagues with affiliations around Eastern Europe, published a highly technical paper.

It describes the development and evaluation of an automated classification method that uses deep learning techniques to analyze the R-R intervals, the time duration between peaks of two successive “R” waves on an electrocardiogram (ECG), which represents the time between heartbeats, measured by a wearable heart monitor, similar to a fitness tracker, that sells for $104.95 at Walmart.

The low cost of the device, the researchers write, along with high reliability and the relatively short time periods that it needs to be worn to capture the necessary data, would make it useful beyond a clinical setting: at home and in rural areas and low-income countries where there are not enough psychiatrists to diagnose mental disorders based on physician interviews.

It also could be used for screening for psychiatric disorders on a population basis that could perhaps lead to faster diagnoses in earlier stages. Disease treatment and progression could be monitored via telemedicine. Unlike many advanced medical diagnostic technologies, highly specialized personnel would not be needed on-site, although the data require contextual interpretation through software or a trained physician.

The study compared multiple machine learning models used to analyze the heartbeat data and published results for each. Diagnostic accuracy rates are based on differences in that data between the schizophrenia/bipolar group and the controls.

The authors discussed at length one potentially significant caveat. Multiple studies have found that antipsychotic medications can significantly influence the heart rate variability that the study relied on, with different drugs having different effects. Yet the researchers could not ethically stop their subjects’ medication.

To assess the possible effect of antipsychotic use specifically on the heartbeat intervals used in the study, the researchers compared those values for a subset of patients receiving quetiapine (Seroquel), which is known to have the largest effect, against the rest of the schizophrenia/bipolar group. No significant differences were found, suggesting that the antipsychotics’ influence on their measures was small.

“Our study indicates that wearable devices may be a cost-effective diagnostic tool, both for in- and outpatients,” the authors conclude.