Your face might reveal what your answers cannot
Future breakthroughs in mental health research could come from combining self-reported experiences with objective measures such as facial movements, voice patterns, and other digital biomarkers. Associate Professor at the IT University of Copenhagen, Stella Grasshof, investigates how to use biometric datasets to gain more insights about mental illnesses.
Researchhealth
Written 25 September, 2026 13:11 by Jari Kickbusch
Patients complete questionnaires measuring depression, anxiety, mania, or quality of life. Researchers analyse the responses statistically and search for patterns. This approach has been invaluable and will continue to play an important role. But if psychology is to move toward more precise diagnostics and more personalized treatment, questionnaires alone are not enough.
We need to complement them with objective biometric data, explains Associate Professor at the IT University of Copenhagen, Stella Grasshof. She is one of the researchers behind a recently published study that shows that facial micro-movements can be used in an interpretable machine-learning framework to distinguish people with bipolar disorder (BD) from healthy controls.
"The results of this study, together with related research, highlight the potential of facial expressions beyond emotions as additional source of information in clinical assessment. We do not aim to replace existing assessments but wish to complement self-reporting and clinical observations by providing practitioners with fine-grained and quantifiable information about patients’ behaviour," she says.
The Problem with Self-Report
Most psychiatric assessments rely on people describing their own thoughts, feelings, and symptoms. This makes intuitive sense because mental illness is fundamentally tied to subjective experience. Yet self-report methods have well-known limitations.
First, individuals vary in their ability to accurately describe their mental state. Some underestimate symptoms, while others overestimate them. Second, responses are influenced by memory, social desirability, and situational factors. The way a person answers a questionnaire on a Tuesday morning may differ significantly from how they respond on a Friday afternoon.
"Surveys primarily tell us what people are aware about which does not necessarily reflect their whole experience. Especially, self-reported surveys heavily rely on the ability to understand and interpret questions as intended and accurately report their own experience. Therefore, observable criteria such as facial expressions, eye movement, behaviour, and speech are commonly used to assess patients, e.g. limited eye contact is an established item on checklists for autism. However, the evaluation of such indicators often requires trained practitioners and depends on the assessor’s expertise, training, and interpretation, introducing inter-rater variability," Stella Grasshof says.
Interdisciplinary approach
In the future, automated analysis could make insights more accessible and interpretable, while supporting clinicians in assessment and potentially helping to track changes over time.
Stella Grasshof doesn't think that diagnosis only by facial expressions is feasible or advisable, but she is certain that biometric data has the potential become highly valuable as an additional source of objective information. However, the several challenges that researchers need to overcome, before biometric data can be used as a supporting tool in the healthcare sector.
"One of the challenges are the different focuses and approaches in the different research areas. In the machine learning community, for example, researchers tend to hunt for new methods and beating the benchmark by some number. The majority do not concern themselves with interpretability or how valuable the chosen quality metric is for clinical use-cases. In the clinical community, researchers do not care about higher accuracy if it is unclear how the number was achieved. Interpretability is responsibility and we need an interdisciplinary approach to find common ground and a way forward," Stella Grasshof ends.
Learn more
Read the article Don’t predict if you cannot interpret: investigating the clinical viability of facial movements for machine-learning assisted diagnostics of bipolar disorder. The article is authored by Martin Lund Trinhammer, Stella Grasshof, Lars Vedel Kessing, Hanne Lie Kjærstad, Kamilla Woznica Miskowiak and Sami S. Brandt.