AI finds a hidden stress signal inside routine CT scans

Researchers have used a deep learning artificial intelligence model to identify what they describe as the first biomarker of chronic stress that can be directly seen on standard medical images. The findings are being presented next week at the annual meeting of the Radiological Society of North America (RSNA).

Chronic stress does not just affect mood. It can influence both physical and mental health, contributing to problems such as anxiety, trouble sleeping, muscle pain, high blood pressure and a less effective immune system, according to the American Psychological Association. Studies have also linked ongoing stress to major conditions including heart disease, depression and obesity.

AI measures adrenal glands on routine CT scans

The study's lead author, Elena Ghotbi, M.D., a postdoctoral research fellow at Johns Hopkins University School of Medicine in Baltimore, Maryland, created and trained a deep learning tool designed to calculate the size of the adrenal glands using CT scans that had already been performed.

Each year, tens of millions of chest CT scans are performed in the United States alone.

"Our approach leverages widely available imaging data and opens the door to large-scale evaluations of the biological impact of chronic stress across a range of conditions using existing chest CT scans," Dr. Ghotbi said. "This AI-driven biomarker has the potential to enhance cardiovascular risk stratification and guide preventive care without additional testing or radiation."

Making the burden of stress visible in the body

Senior author Shadpour Demehri, M.D., professor of radiology at Johns Hopkins, noted that chronic stress is extremely common and is something many adults experience every day.

"For the first time, we can 'see' the long-term burden of stress inside the body, using a scan that patients already get every day in hospitals across the country. Until now, we haven't had a way to measure and quantify the cumulative effects of chronic stress, other than questionnaires, surrogate serum markers like chronic inflammation, and cortisol measurement, which is very cumbersome to obtain." Dr. Demehri said.

Unlike a single cortisol test, which reflects stress at just one point in time, the size of the adrenal glands functions more like a long-term gauge of chronic stress.

Large multi-ethnic cohort links imaging, hormones and stress load

In this research, the team analyzed information from 2,842 participants (mean age 69.3; 51% women) enrolled in the Multi-Ethnic Study of Atherosclerosis, a large study that combines chest CT imaging, validated stress questionnaires, cortisol measurements and indicators of allostatic load -- the cumulative physiological and psychological effects of chronic stress on the body. Because it integrates imaging, biochemical data and psychosocial assessments in the same individuals, this cohort was uniquely suited, and likely the only one available, for creating an imaging-based marker of chronic stress.

The investigators applied their deep learning model to the CT scans to automatically outline and measure adrenal gland volume. They defined Adrenal Volume Index (AVI) as adrenal volume (cm3) divided by height2 (m2). To capture hormonal patterns, participants provided salivary cortisol eight times per day over the course of two days. Allostatic load was calculated using body mass index, creatinine, hemoglobin, albumin, glucose, white blood count, heart rate and blood pressure.

Adrenal Volume Index tracks stress, hormones and heart risk

The team then examined how AVI related to cortisol, allostatic load and a range of psychosocial stress indicators, such as depression scores and perceived stress questionnaires. They discovered that AVI generated by the AI model aligned with established stress questionnaires, with circulating cortisol levels and with future adverse cardiovascular events.

Higher AVI values were linked with greater overall cortisol exposure, higher peak cortisol levels and increased allostatic load. People who reported high levels of perceived stress had higher AVI compared with those who reported low stress. AVI was also connected to a higher left ventricular mass index, a measure related to heart structure. For every 1 cm3/m2 increase in AVI, the risk of heart failure and death increased.

"With up to 10-year follow-up data on our participants, we were able to correlate AI-derived AVI with clinically meaningful and relevant outcomes," Dr. Ghotbi said. "This is the very first imaging marker of chronic stress that has been validated and shown to have an independent impact on a cardiovascular outcome, namely, heart failure."

A new way to quantify the cumulative impact of stress

"For over three decades, we've known that chronic stress can wear down the body across multiple systems," said Teresa E. Seeman, Ph.D., study co-author and professor of epidemiology at UCLA and a pioneering researcher in stress and health. "What makes this work so exciting is that it links a routinely obtained imaging feature, adrenal volume, with validated biological and psychological measures of stress and shows that it independently predicts a major clinical outcome. It's a true step forward in operationalizing the cumulative impact of stress on health."

Dr. Demehri explained that connecting a simple imaging measure with several well-established markers of stress and disease outcomes creates a new, practical approach to measuring chronic stress in everyday clinical practice.

"The key significance of this work is that this biomarker is obtainable from CTs that are performed widely in United States for various reasons," Dr. Demehri said. "Secondly, it is a physiologically sound measure of adrenal volume, which is part of the chronic stress physiologic cascade."

The researchers noted that this imaging biomarker could potentially be applied to many stress-related diseases that commonly affect middle-aged and older adults.

Other co-authors are Roham Hadidchi, Seyedhouman Seyedekrami, Quincy A. Hathaway, M.D., Ph.D., Michael Bancks, Nikhil Subhas, Matthew J. Budoff, M.D., David A. Bluemke, M.D., Ph.D., R. Graham Barr and Joao A.C. Lima, M.D.