Making the Case for CT-Based Sarcopenia Screening
How deep learning and large-scale CT analysis can make muscle loss measurable, actionable and easier to diagnose
Sarcopenia is an age-related loss of muscle mass and function that can contribute to frailty, disability and worse clinical outcomes. Although it affects between 5% and 10% of the population, according to Richard Zhuang, an undergraduate research assistant at Penn Medicine, the condition remains underrecognized and inconsistently diagnosed in clinical practice.
“Even though sarcopenia now has an ICD-10 code, there is still no universally accepted diagnostic threshold, and many clinicians are unfamiliar with how to measure it,” Zhuang said.
A new Penn Medicine study aims to change that.
“By analyzing CT-derived muscle and fat phenotypes alongside clinical, laboratory and genetic data, we wanted to understand how muscle loss and fat redistribution occur across a large population and whether these imaging markers can improve prediction and diagnosis,” said Zhuang, who discussed the study at RSNA 2025.
Leveraging deep learning algorithms that automatically segment muscle and fat, Zhuang and colleagues looked at abdominal CT scans from more than 7,500 patients in the Penn Medicine BioBank. From these scans, the researchers quantified total abdominal muscle volume and visceral, subcutaneous and intramuscular fat.
Having evaluated how these body composition measurements varied with age and sex, the study then used the distributions to estimate potential diagnostic thresholds for sarcopenia.
Beyond imaging trends, the study also linked its measurements to clinical data. Specifically, a phenome-wide association study was performed to identify diseases associated with muscle loss and fat infiltration. This was followed by a genome-wide association study exploring the genetic variants that contribute to these traits.
“Our approach allowed us to connect imaging-derived body composition metrics with broader clinical and biological factors related to sarcopenia,” Zhuang explained.
The Many Benefits of CT-Based Body Composition Metrics
One of the study’s key findings was how muscle mass and fat content follow predicable trajectories. While the former slightly increases into midlife, peaking during the fifth decade of life and then steadily declining with age, intramuscular and visceral fat continuously increases with age.
“This reflects the gradual replacement of muscle with fat that characterizes sarcopenia,” Zhuang remarked.
The team also noted how body composition measurements are associated with such metabolic laboratory markers as HbA1c, triglycerides and HDL cholesterol, reinforcing the link between muscle composition and metabolic health. “Because CT can directly quantify muscle volume and fat infiltration, imaging can serve as a noninvasive biomarker of these metabolic changes,” Zhuang said.
Other notable findings include an association between imaging metrics and a range of metabolic and clinical conditions, including diabetes, hypertension and cardiovascular disease. The study further identified several loci potentially associated with muscle volume and fat infiltration.
“These findings establish population-level reference patterns for muscle and fat distribution and suggest that CT-based body composition metrics can help identify patients at risk for sarcopenia and related diseases,” Zhuang explained.
He noted how the incidental detection of sarcopenia can also serve as a call-to-action finding that prompts further evaluation or intervention, especially in patients at risk for frailty or metabolic disease.
“There is growing interest in incorporating body composition metrics into radiology reports, particularly if they provide actionable clinical information,” he said. “For sarcopenia, this might eventually involve identifying whether a patient falls below established muscle volume thresholds or shows significant fat infiltration within muscle.”
Zhuang noted that incorporating these metrics into routine reporting could enable radiologists to flag sarcopenia earlier and provide clinicians with quantitative data to guide risk stratification and treatment planning. Standardized muscle volume thresholds and fat infiltration markers may also support more consistent assessment across imaging practices.
Accuracy and Automation: A Powerful Combination
The Penn Medicine study clears a path towards using routine CT for sarcopenia screening.
“Because abdominopelvic CT scans are already regularly performed for a range of clinical indications, for many patients, we already have the imaging data needed to assess muscle and fat composition,” Zhuang noted. “Now radiologists only need the time to measure these structures.”
This is where AI comes into play.
With automated AI segmentation, radiologists no longer need to do these measurements manually. In fact, they could potentially be generated in the background without adding to the radiologist’s workload. They could also be used to automatically flag any abnormal body composition metrics, allowing radiologists to easily include a concise statement or recommendation in their report.
“AI makes it possible to automate this entire process, enabling rapid and consistent segmentation of muscle and fat across large imaging datasets,” Zhuang said. “This combination of accuracy and automation is what makes CT-based body composition analysis particularly powerful.”
“AI makes it possible to automate this entire process, enabling rapid and consistent segmentation of muscle and fat across large imaging datasets. This combination of accuracy and automation is what makes CT-based body composition analysis particularly powerful.”
— RICHARD ZHUANG
Turning CT Body Metrics into Earlier, Equitable Care
By establishing body composition trends in a large patient population, the Penn Medicine study establishes reference points that clinicians can now use to evaluate whether an individual patient is at an elevated risk for metabolic disease, frailty or other adverse outcomes.
In the long term, automated body composition analysis could take this one step further, helping identify at-risk patients earlier and guide interventions such as nutritional counseling, exercise programs or metabolic management.
“Allowing researchers to better understand how muscle loss and fat redistribution contribute to chronic disease across large patient cohorts, our study helps bridge the gap between accurately evaluating risk and effectively managing it,” Zhuang said.
But as these AI-enabled risk insights move from research into routine care, they also raise questions about how the information might be interpreted, and by whom.
Though his team’s research did not specifically examine health insurance implications, Zhuang noted that it is a question that deserves serious consideration as AI-driven imaging tools become increasingly integrated into clinical practice.
By identifying individuals at risk for sarcopenia earlier, Zhuang and his team hope to enable timely interventions that improve quality of life and reduce the burden of downstream complications such as falls, fractures and hospitalizations.
“As with any tool that enhances our ability to stratify health risk, there is a responsibility to ensure that these insights are used to help patients rather than disadvantage them,” he said. “The broader health care community must work together to ensure that advances in precision medicine translate into equitable care.”
For More Information
Access educational content related to sarcopenia on EdCentral.
Read previous RSNA News stories on sarcopenia: