A team of researchers has demonstrated that an artificial intelligence model can now detect physiological signals associated with chronic stress using routine medical imaging scans. This breakthrough, presented at the annual meeting of the Radiological Society of North America (RSNA), marks a significant step forward in the search for an objective biomarker of prolonged stress, a phenomenon whose clinical impact is well documented but still difficult to quantify in a standardized manner1.
The study is based on an analysis of chest CT scans performed on nearly 3,000 patients. The researchers applied a deep learning model capable of automatically extracting structural information about the adrenal glands, which are essential organs involved in the hormonal response to stress. The measurements obtained were combined with various clinical data, ranging from cortisol levels to blood pressure, in order to identify a robust and observable physiological marker reflecting the burden caused by chronic stress.
A biomarker detected through automated analysis of the adrenal glands
The adrenal glands play a central role in regulating metabolism, blood pressure, the immune system, and stress responses. In this study, they serve as the primary biological source of the detected biomarker. Patients with high stress levels prior to the examination had, on average, larger adrenal glands, elevated cortisol levels, and a higher risk of heart failure2.
This finding represents a significant contribution. Until now, chronic stress has been assessed using questionnaires or one-time hormonal measurements—methods that are useful but limited. The ability to identify a physiological marker directly in routine medical images opens up a new avenue for standardized and reproducible screening.
A method based on the analysis of thousands of medical images
To achieve this result, the researchers built a deep learning model trained on thousands of chest CT scans. They automatically extracted the size of the adrenal glands and then compared this data to various clinical indicators associated with chronic stress.
Some key figures from the study:
- nearly 3,000 patients included,
- a large number of CT scans analyzed,
- a model trained to recognize subtle variations in adrenal morphology,
- strong correlations between adrenal size, cortisol, and cardiovascular risk.
This ability to extract biomedical signals invisible to the human eye confirms the potential of AI in the early detection of markers of mental and physical health.
Medical Applications: Toward a More Objective Diagnosis of Chronic Stress
The identification of a biomarker for chronic stress opens up significant opportunities for prevention and medical management.
Among the potential applications:
- early detection of cardiovascular risks associated with prolonged stress,
- longitudinal follow-up of patients exposed to stressful environments,
- objective assessment of the effectiveness of therapies,
- development of prevention tools in the workplace,
- modeling stress vulnerability profiles to tailor care pathways.
This type of biomarker could also contribute to more targeted predictive medicine by identifying individuals at risk of physiological conditions that do not manifest in their symptoms.
Scientific limitations and precautions
Like any preliminary findings, these results should be interpreted with caution. The study has not yet been published in a peer-reviewed journal, and the widespread adoption of the biomarker requires:
- validation in more diverse cohorts,
- greater statistical rigor,
- the adjustment for confounding factors (age, physical condition, comorbidities),
- longitudinal analyses to confirm the marker's stability over time.
While adrenal volume is linked to chronic stress, it cannot, on its own, fully capture the complexity of the phenomenon.
Ethical Issues: Sensitive Data and Risks of Non-Medical Use
This discovery also raises significant ethical issues. Medical images containing physiological indicators related to stress become particularly sensitive data. Their use in non-medical contexts—for example, for performance evaluations or insurance underwriting—would constitute a major misuse.
Health data governance, informed consent, and restrictions on non-clinical uses are therefore essential to support the emergence of these new tools. Future regulations, particularly at the European level, will need to incorporate these new technological capabilities in order to prevent inequalities and discrimination.
Conclusion: A Turning Point for Mental Health and Personalized Medicine
The identification of a biomarker for chronic stress using AI represents a significant breakthrough in understanding the physiological effects of stress. It paves the way for more objective diagnosis, more targeted prevention, and more personalized care.
This discovery shows that AI, when combined with medical imaging, can reveal physiological signals that were previously undetectable. It is part of a broader trend in which medicine is leveraging automated analysis to develop predictive tools and facilitate the early identification of mental health risks.
Learn more
To learn more about the evolution of artificial intelligence tools in healthcare and understand how these technologies are reshaping medical practices, see: MedGPT: The Free French Medical AI That Rivals ChatGPT
References
1. Radiological Society of North America. (2025). Deep Learning Model Identifies Adrenal Biomarkers of Chronic Stress From Routine CT Imaging.
https://www.rsna.org
2. National Institutes of Health. (2024). Physiological Mechanisms of Chronic Stress: Cortisol, Adrenal Function, and Long-Term Health Outcomes.
https://www.nih.gov
