Perry Pickhardt, MD, Professor of Radiology and UWSMPH faculty member in the Abdominal Imaging and Intervention Section, and National Institutes of Health researchers, recently published their findings that demonstrate artificial intelligence (AI) can harness and analyze information captured in computerized tomography (CT) scans, but typically not used, to provide valuable medical information such as predicting future adverse clinical events.
Their study compared the ability of automated CT-based body composition biomarkers derived from deep-learning and feature-based, image-processing algorithms (previously developed at the NIH Clinical Center) to predict major cardiovascular events and overall survival with routinely used clinical parameters: the Framingham risk score (FRS) and body-mass index (BMI). The investigators observed that the CT-based measures were more accurate than FRS and BMI in predicting downstream adverse events including death or myocardial infarction, cerebrovascular accident, or congestive heart failure subsequent to CT scanning in a cohort of adults undergoing abdominal CT for routine colorectal cancer screening. The results appeared in The Lancet Digital Health on March 2, 2020.
“This opportunistic use of additional CT-based biomarkers provides objective value to what doctors are already doing,” said Dr. Perry J. Pickhardt, professor of radiology at the University of Wisconsin School of Medicine & Public Health, and lead and corresponding author of the study. “More than 80 million body CT scans are performed every year in the U.S. alone, but valuable prognostic information on body composition is typically overlooked. This automated process requires no additional time, effort, or radiation exposure to patients, yet these prognostic measures could one day impact patient health through pre-symptomatic detection of elevated cardiovascular or other health risks.”
The study used five AI computer programs to accurately measure liver volume and steatosis, visceral fat volume, skeletal muscle volume, spine bone mineral density, and atherosclerotic plaque burden on abdominal CT examinations, and they found that the panel of automated CT-based tissue biomarkers compared favorably with the FRS and BMI for presymptomatic prediction of future cardiovascular events and death. In fact, the CT-based measure of aortic calcification alone significantly outperformed the multivariate FRS for major cardiovascular events and overall survival.
The researchers also observed that BMI was a poor predictor of cardiovascular events and overall survival, and all five automated CT-based measures clearly outperformed BMI for adverse event prediction.
“We found that automated measures provided more accurate risk assessments than established clinical biomarkers,” said Dr. Ronald M. Summers of the NIH Clinical Center and senior author of the study. “This demonstrates the potential of an approach that uses AI to tap into the biometric data embedded in all such scans performed for a wide range of other indications and derive information that can help people better understand their overall health and risks of serious adverse events.”
This research builds on prior efforts designing AI algorithms that Dr. Summers has undertaken in his lab in the NIH Clinical Center’s Radiology and Imaging Sciences Department and his previous collaboration with Dr. Pickhardt to develop, train, test, and validate fully automated algorithms for measuring body composition using abdominal CT. The researchers plan to test the approach in other cohorts, including more racially diverse populations.
This study was supported by the Intramural Research Program, NIH’s internal research program and the world’s largest biomedical research institution, and it used the high-performance computing capabilities of the NIH Biowulf cluster.
Pickhardt P, Summers R, et al. Automated CT biomarkers for opportunistic prediction of future cardiovascular events and mortality in an asymptomatic screening population: a retrospective cohort study. The Lancet Digital Health. March 2, 2020. DOI: 10.1016/S2589-7500(20)30025-X