The following article is from the Advanced Heart Failure referring provider newsletter.
A new study, led by Veli Topkara, MD, medical director of the Heart Failure, Cardiac Transplantation and Mechanical Circulatory Support Program at UW Health, has shown that artificial intelligence (AI) may help predict who will do best after a heart transplant.
It is the first study that uses deep learning to automatically measure frailty on computed tomography (CT) scans and link it to heart transplant survival. While earlier studies have hinted at the importance of muscle mass, they relied on manual measurements. Automating the process removes a major barrier to incorporating sarcopenia assessment into daily practice.
“This tool opens the door for more personalized transplant care,” Dr. Topkara said. “By identifying high-risk patients early, we can intervene in ways that may improve sarcopenia, survival and quality of life of our patients.”
Sarcopenia and frailty
While sarcopenia is often associated with aging, it also develops in people with chronic illness, including advanced heart failure. In these patients, muscle wasting reflects frailty, lower physical reserves and a body under stress.
“Frailty is increasingly recognized as an important factor influencing outcomes after heart transplantation,” said Dr. Topkara. “Patients with more muscle mass tend to recover faster, while those with sarcopenia are more vulnerable to complications.”
Until now, assessing muscle mass has been difficult. The best method involves manually tracing areas of muscle on CT scans, a process that is slow, costly and prone to variation between doctors. This makes it impractical for routine use in evaluating transplant candidates. Additionally, few transplant centers have a trained radiologist who is able to provide this service on a regular basis.
AI-assisted diagnosis
Dr. Ronald Summers, a pioneer in medical imaging and AI at the National Institutes of Health (NIH), developed an AI tool that takes just seconds to determine a patient’s muscle mass and, therefore, their frailty. He and Dr. Topkara used this tool in their study to make frailty measurement an entirely automated process.
For this project, the researchers applied a deep learning algorithm to chest CT scans from 164 adults who underwent heart transplantation between 2013 and 2022.
The AI tool automatically measured muscle area at specific spinal levels and calculated a skeletal muscle index (Figure 1).
Study findings
The results revealed a strong link between sarcopenia and post-transplant outcomes:
Survival
At one year, survival was 98% in patients without sarcopenia versus 90% in those with it. At three years, survival was 97% versus 83%, respectively.
Risk
After adjusting for other factors, sarcopenia nearly quadrupled the risk of graft failure or death.
Hospital stay
Patients with sarcopenia spent longer in the hospital after transplant—28 days on average versus 20 days for others.
By identifying sarcopenia before surgery, doctors could anticipate which patients might need added support or targeted rehabilitation to improve recovery. They also could determine whether patients would benefit from physical therapy before transplantation.
“With some patients, instead of transplant, we could offer them an LVAD (left ventricular assist device) and let them get stronger and improve their sarcopenia,” said Dr. Topkara.
Implications
If AI-based sarcopenia assessment becomes adopted widely, it could:
Streamline evaluations:
CT scans are already part of many transplant evaluations, so AI analysis adds value without additional testing.
Guide personalized care:
At-risk patients could receive nutritional support, physical therapy or closer monitoring.
Reduce costs:
Automated analysis saves time and avoids variability between human readers.
Enable population-level screening:
Hospitals could assess sarcopenia across large groups, something previously unrealistic.
Next steps
The researchers believe this is only the beginning. Future studies may explore additional AI-derived imaging biomarkers beyond muscle mass. Subtle features in CT scans may hold untapped predictive power that AI can uncover.
Dr. Topkara is establishing a heart failure AI research program at UW Health. The program will focus on developing and applying artificial intelligence tools to improve care for patients with advanced heart failure, bridging cardiology and data science in new ways.
“Our vision is to build a research hub at UW Health that leverages AI to transform how we evaluate, treat and follow patients with heart failure,” said Dr. Topkara. “This study is just the first step toward that larger goal.”
As AI evolves, its role in medicine is expected to expand rapidly. For patients facing the daunting journey of heart transplantation, these advances could mean more tailored care, smoother recoveries and more years of life with a healthy heart.
Topkara, et al. Sarcopenia Assessment Using Fully Automated Deep Learning Predicts Cardiac Allograft Survival in Heart Transplant Recipients. Circulation: Heart Failure.